CA2265060C - Word grouping accuracy value generation - Google Patents
Word grouping accuracy value generation Download PDFInfo
- Publication number
- CA2265060C CA2265060C CA002265060A CA2265060A CA2265060C CA 2265060 C CA2265060 C CA 2265060C CA 002265060 A CA002265060 A CA 002265060A CA 2265060 A CA2265060 A CA 2265060A CA 2265060 C CA2265060 C CA 2265060C
- Authority
- CA
- Canada
- Prior art keywords
- word
- accuracy
- character
- value
- words
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/12—Detection or correction of errors, e.g. by rescanning the pattern
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/26—Techniques for post-processing, e.g. correcting the recognition result
- G06V30/262—Techniques for post-processing, e.g. correcting the recognition result using context analysis, e.g. lexical, syntactic or semantic context
- G06V30/274—Syntactic or semantic context, e.g. balancing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Abstract
The present invention is a computer-implemented method for calculating word accuracy. Word grouping accuracy values (260) are calculated (212) by using the character accuracy values (250) calculated by an OCR program present in a computer system. The present invention preferably uses these character accuracy values (250) to create a word grouping accuracy value (260). Various methods are employed to calculate the word accuracy (260), including binarizing the character accuracy values (250). modified averaging of the character accuracy values (250), and creating fuzzy visual displays of word grouping accuracy values (260). The calculated word grouping accuracy values (260) are then adjusted based upon known OCR strengths and weaknesses. and based upon comparisons to stored word lists and the application of language rules. In a system with multiple character recognition techniques. the system can compare the accuracy values (260) of different versions of the word groupings to find the most accurate version. Then, the most accurate version of the word groupings is kept.
Description
Word Grouping Accuracv Value Generation Technical Field This invention pertains to the field of data storage and filing systems, more specitically, to those systems emploving optical character recognition.
Background Art The field ot'document imaging is growing rapidiv, as modern society becomes more and more digital. Documents are stored in digital format on databases, providing instantaneous access, minimal phvsical storage space, and secure storage.
Todav's society now faces questions on how best to transfer its paper documents into the digital medium.
The most popular method of digitizing paper documents involves using a system comprising a scanner and a computer. The paper documents are fed into a scanner. which 1 -S creates a bitmap image of the paper document. This bitmap image is then stored in the computer. The computer can take a varietv of forms. including a single personal computer (PC) or a network of computers using a central storage device. The bitmapped images must be able to be retrieved after they are stored. One system for filing and retrieving documents provides a user interface which allows a user to tvpe in a search 10 term to retrieve documents containing the search term. Preferablv, the svstem allows the user to type in any word that the user remembers is contained within the desired document to retrieve the desired document. 1{owever, in order to retrieve docunients on this basis. the document must be character recognized. That is. the computer must recognize characters within the bitmapped image created by the scanner.
Another common usage of digitizing documents is to digitize long paper documents in order to allow the document to be text searched by the computer.
In this usage, a user types in the key word the user is looking for within the document, and the system must match the search term with words found within the document. For these systems. the document must be character recognized as well.
The most common method of recognizing characters is by using an optical character recognition (OCR) technique. An optical character recognition technique extracts character information from the bitmapped image. There are many different types of optical character recognition techniques. Each has its own strengths and weaknesses.
For example. OCR I may recognize handwriting particularly accurately. OCR 2 may recognize the Courier font well. If OCR I is used to recognize a document in Courier font, it mav still recognize the majority of the characters in the document.
However. it may recognize many of the characters inaccurately. A user may not know of an OCR's strengths and weaknesses. A user may not know whether or not the types of documents the user typicallv generates are of the kind that are accurately recognized by the OCR
present on the user's system. Current systems do not inform the user of the quality of the recognition of the OCR technique. The user finds out how accurate the recognition was only by using the document for the purpose for which it was stored into the computer system, at which time it may be too late to correct.
I 9100-3177/ /7091 13.1 An inaccuratelv recounized document can lead to several problems. First of all, in a system in which documents are stored and retrieved based on their contents.
an inaccurately recognized document may become impossible to retrieve. For example. if a user believes the word "imaging" is in a specitic document. the user will type in "imaging" as the search term. However, if the word "imaging" is recognized incorrectly.
such that it was recognized as "emerging." the user's search %vill not retrieve the desired document. The user may not remember anv other words in the document. and thus the document is unretrievable. In a svstem xhere documents are dittitized to allow text searching of the document, the same problem occurs. Misrecognized words are not tound by the use of the correct search terms.
Thus, there is a need to allow the user to determine whether a recognized word is of acceptable quality. By allowing the user to determine whether akvord is of acceptable quality. the user can ensure that the document is retrieved by the use of that %vord as a search term. Also. a user can ensure that words within the document are accurately recognized for internal document searching purposes. Additionallv. in a system . ith multiple optical character recognition techniques. there is a need to be able to compare the accuracy of the different versions of the document to create a version that is the most accurate.
Disclosure of the Invention The present invention is a computer-implemented method lor calculating \~ord groupinu accuracy values (260). l'he present invention receives (200) data.
pertorms (204) an optical character recoi-nition technique upon the received data. and creates (208) I9I00-3177 'O9113.1 word groupings. The system then calculates (212) word grouping accuracy values (260) for the created %vord groupings.
Word grouping accuracy values (260) are calculated (212) by using character accuracy values (250) determined by the OCR technique. The present invention preterably uses these character accuracy values (250) to create a word grouping accuracy value (260). Various methods are employed to calculate the word accuracy (260), including binarizing the character accuracy values (250), modified averaging of the character accuracy values (250), and employing fuzzy visual displays of word grouping accuracv values (260). The calculated word grouping accuracy values (260) are adjusted based upon known OCR strengths and weaknesses, and based upon comparisons to stored word lists and the application of language rules. Word grouping accuracy values (260) are normalized and displayed or compared to a threshold. The words whose accuracy values (260) exceed the threshold may then be used to index the documents or provide search terms tor searching within the document. If no word groupings exceed the threshold then the user is offered different options, including to clean the image by performing another OCR or scanning the document again, or to reset the threshold to a lower value.
Brief Description of the Drawinps -)0 These and other more detailed and specific objects and features of the present invention are more fully disclosed in the following specification, reference being had to the accompanying drawings. in which:
19100-3177//709113.1 Figure 1 is a block diagram of a computer system embodying the present invention.
Figure 2a is a flowchart diagram illustrating the process steps of an embodiment of the present invention in which word accuracy is calculated for a single OCR
system.
~ Figure 2b is an illustration of two characters and their accuracy values 250.
Figure 2c is an illustration of a word grouping and its individual character accuracy values 250.
Figure 3 is a more detailed flowchart of process step 208 of the present invention.
Figure 4 is a more detailed flowchart of process step 212 of the present invention.
Figure 5a is a more detailed flowchart illustrating the assigning binary representation module 412 of the present invention.
Figure 5b is an illustration of word groupings and accuracy values 260 in accordance with the present invention.
Figure 5c is a flowchart of one alternate embodiment of calculating word grouping accuracy values 260 in accordance with the present invention.
Figure 5d is a flowchart of a second alternate embodiment of calculating word grouping accuracy values 260 in accordance with the present invention.
Figure 6a is one embodiment of the adjusting word accuracy module 416 of the present invention.
Figure 6b is an alternate embodiment of the adjusting word accuracy module 416 of the present invention.
19100-3177'1709113.1 Figure 7 is a flowchart diagram illustrating the process steps of an embodiment of the present invention in which word accuracy is calculated for a multiple OCR
system.
Fieure 8a is a more detailed flowchart of process step 724 of the present invention.
Figure 8b is an illustration of a word grouping version table 850.
Figure 9 is a more detailed flowchart of process step 728 of the present invention.
Figure 10 is a more detailed flowchart of process step 736 of the present invention.
Figure 1 I is a flowchart illustrating an updating indexing list embodiment in accordance with the present invention.
Detailed Description of the Preferred Embodiments Figure 1 illustrates a hardware embodiment of the present invention. A data receiver 100 is coupled to the data bus 110 of a computer 120. The data receiver 100 preferably receives data from a scanner 104, which can be any conventional scanner, such as Canon DR 3020 or Canon PICS 4400. Alternatively, the data receiver can be a network adapter or [nternet connection comprising a modem and a phone line which couples the computer 120 to a remote data source from which data is received.
The computer 120 includes a central processing unit 130, such as an Intel Pentium processor, a displav 140. an input device 190, random access memorv (RAM) 150, read-onlv memory (ROM) 155, and a disk 160, all interconnected through bus 1 10.
1 9 1 00-3 1 77 .'7091 13.1 txn input device 190 is a mouse. keyboard. or other pointine device that allows a user 180 to interact Nk-ith the coniputer 120. ROM 155 provides CPU 130 -with unvarying tunctions such as executine programs initialized at the start-up otcomputer 120. R.AM
150 provides the CPU 130 with data and instruction sequences that are loaded trom disk 160. The disk 160 is. for example, 1.6 Gigabyte disk that intertaces to the computer bus throui-yh a Small Computer Svstem Interface (SCSI) interface 165. Disk 160 stores data tiles such as binarv. grav-scale. and color image tiles, a table 170. text tiles, and proerams includine a word accuracv calculation program and one or more OCR programs.
Disk 160 is a single device or multiple devices. The components of computer 120 take conventional form. Computer 120 can stand alone or is a part of a network of computers 125 comprised of individual computers 120 coupled together through Ethernet connections. The present invention can be used in conjunction with any computer operating system. such as Windows 95. Windows NT. or OS/2, resident in memorv 155.
As illustrated in Figure 2. the present invention receives 200 data. As described above, data can 200 be received in a multitude ot wavs. including by a scanner or hv a remote connection with another computer. I f a scanner 104 is coupled to data recei% er 100, then the paper document fed into the scanner 100 is digitized. Digitizing a paper document involves creating a bitmap of the paper document. Atter a digitized version of the document is created, an OCR is pertormed 204 upon the bitmap. A Canon OCR
technique. specit7ed in U.S. 11'at. 5.379,349, is preferably used as the OCR
technique:
however. any OCR technique can he used in accordance %vith the present invention. The OCR technique creates character information trom the bitmap. :\ description othow the character recognition process is accomplished is found in the above-mentioned U.S. Pat. 5,379,349. A document may be a single piece of paper or multiple pieces of paper that are substantially related.
Most OCRs provide a character accuracy value 250 for characters recognized from the bitmap. The character accuracy value 250 is typically created by comparing the recognized character with a stored template of the character.
The differences in the two versions of the character are quantized and represented as an accuracy value 250. For example, a recognized "p" and a stored template of a "p" are shown in Figure 2b. This "p" is given a character accuracy value 250 of .95 because of the differences in shape between the recognized "p" and the stored template of the "p".
The system next creates 208 word groupings from the recognized characters, as illustrated by process steps 300, 304, 308, 312 and 316 shown in Figure 3.
After word groupings are created 208 from the extracted character information of the bitmap of the paper document, the system calculates 212 the word grouping accuracy values 260 of the created word groupings. Word grouping accuracy values 260 are calculated to provide the user 180 with an easily understood measure of the quality of an OCR's recognition. Displaying a string of character accuracy values 250 may not communicate useful information to the average user 180. For example, in Figure 2c, the word "rock"
is composed of four characters "r", "o", "c", and "k." The "r" is recognized at a .95 accuracy value 250, the "o" is recognized at a .72 accuracy value 250, the "c" is recognized at a .92 character accuracy value 250 and the "k" is recognized at a .82 accuracy value 250.
Displaying individual character accuracy values 250 does not inform the user whether the Ntiord "rock" %tas accurately recognized. The user 180 knows that "r" and "c"
were accuratelv recognized. but the user 180 does not know how the poor recognition of the "o" and the "k" affects the recognition of the word as a whole.
Determining the accuracy otthe word, as opposed to individual characters, is critical in systems that allow retrieval of documents bv the use of text-searching. In those systems. when the user 180 searches tor a document, the user 180 enters a search tetm composed of words.
Thus, it is more important to know whether an entire word is accurately recognized. as opposed to whether individual characters are accurately recognized. Additionally, in documents in which OCR techniques are performed in order to allow the document itself to be text searched. it is important to know whether the OCR has accurately recognized the document. If the OCR recounition is inaccurate, words searched for will not be tound.
even if the words are present within the image. Finally, if multiple words are displaved with all of their individual character accuracy values 250, the page soon becomes an indecipherable conglomeration of letters and numbers.
Figure 4 illustrates the process steps of an embodiment of the current invention calculating word accuracy in a system employing a single OCR. The svstem selects one ot the newlv created word grouping using a selecting module 400. The svstem then obtains the character accuracv value 250 for one of the characters in the selected word grouping by using an obtaining character accuracv value module. As described above.
1 9100-3 1 77-,'7091 13.1 character accuracy values 250 are typically available after perforrning an OCR. The system then applies a character accuracy value adjusting module 408 to adjust the character accuracy value 250 based upon known OCR weaknesses. This adjustment is based upon the fact that OCRs typically have unique strengths and weaknesses.
The system determines whether the character being examined is one of the known characters that is not accuratelv recognized, or is a font, typeface, or style that is not accurately recognized. In this situation, the system subtracts a constant from the obtained character accuracv value 250. The constant is typically .05, but can be set higher or lower, depending upon user preferences. For OCR strengths, the adjusting module 408 adjusts the character accuracy value 250 upwards to reflect characters, fonts, or typefaces that the OCR typically recognizes more accurately than other recognition systems In the example of Figure 2c, perhaps the word "rock" was created using the Times New Roman font, and the OCR applied to the document recognizes the Times New Roman font accurately. Then. the individual character accuracy values 250 are adjusted upwards to become "r": 1.0, "o": .77. "c": .97, and "k": .87. Thus, although the OCR had a confidence factor of only .72 that the "o" was really an "o". the system of the present invention raises the value to .77 due to the strength of that OCR's ability to recognize Times New Roman. If this number were displayed to the user 180, the user 180 would feel more contident that the "o" was correctly recognized because of the adjustment of the present invention.
Of course. if a document contains words created in one font that is either strongly or weaklv recognized by the OCR present in the system, than all of the characters are 19100-3 I 77,'/7091 13 .1 adjusted uniformlv. and the adjustment is factored out. tlowever. in multiple OCR
situations, discussed belok~. it is important that the adjustment occur. as the different versions of the word groupings created by each OCR are compared against each other.
The above process is repeated tor each character in the selected -ord grouping. and is ~ repeated for each word erouping in the docunlent.
An assigning module 412 is then applied to assign a binary representation to the characters in the selected word grouping. discussed in more detail below. A
preliminary word strouping accuracy value 260 is therebv generated for the selected word grouping.
A word accuracy adjusting module 416 is applied to the selected \,vord grouping to adjust the preliminary word grouping accuracy value 260 based upon predefined tactors such as stored list comparisons or the application of stored language rules. discussed in more detail below. This process is preferably repeated for each word grouping in the received group of data. After a word grouping accuracy value 260 is calculated for the last v.-ord grouping. the system proceeds to displav the word groupings and their accuracv values 260 using a displav module 216. The above-described modules mav be implemented as hard-ware. firmware. or soft%vare, in accordance with the present invention.
Figure 5a illustrates in more detail the assigning binary representation module 412. The svstem selects 500 a character in the selected word grouping. Then.
the system determines 504 whether the tirst character in the word grouping has an accuracy value 1-50 that exceeds a threshold accuracy level. The threshold accuracy level is preferably user detined. The user 180 can theretore set the recognition quality ot the entire s~ stem based upon what threshold the user 180 chooses. For example. it the user 180 desires a I 9101)- 3 177 709113.1 svstem that accepts characters at 90% accuracy, the user 180 can set the threshold at .9.
I f the user 180 is more tolerant of error, the user 180 can set the threshold lower.
The svstem compares the accuracy of the selected character to the user-defined threshold to determine what representation to assign the character. If the character ~ accuracy value 250 exceeds the threshold, the character is assigned 508 a"1." If the character accuracy value 250 does not exceed the threshold, the character is assigned 512 a"0." The system determines 516 if there are more characters. If there are, a counter is incremented 520. and the process is repeated. If all of the characters have been assigned a representation. the system then adds 524 the number of characters assigned a"one"
together to determine the total number of characters assigned a "one". Then the system determines 528 the total number of characters in the word grouping. Finally, the system calculates the preliminary word accuracy for the word grouping by dividing 532 the number of characters assigned a "one" by the total number of characters in the word grouping.
1~ The system uses the above method of calculating because it normalizes word grouping accuracv values 260. Normalizing provides a better representation of the recognition quality of a word. For example, in Figure 5b, the word "ear" has character accuracy values 250 of.93, .89, and .72 assigned to its characters. The threshold is .8.
Theretbre. the "e" and the "a" are assigned a 1 and the "r" is assigned a 0.
The word grouping, accuracy value 260 for "ear" is therefore .67 (2/3). The next word is "earring."
I'he charactcr accuracy values 250 are the same for the first four letters, the "i" is recognized at a.9;. the "n" is recognized at a .85. and the "g" is recognized at a .87.
1 9 1 00-3 1 77//709 1 1 3. t Therefore, its word accuracv is .71 (5/7). The percentage of characters with an accuracy ~ alue above the threshold is greater in the word "earring" than in the word "ear." This is retlected in the -,%ord accuracy value assigned to each word. In one embodiment, repeated characters are discounted for the purposes of calculating word grouping accuracy values ~ 1-60. 1 n the above example, the "r" would only be counted once in calculating the accuracy value, leading to an accuracy value of .83 (5/6). This embodiment is useful when the user 180 is attempting to evaluate the accuracy of the OCR, as the OCR is not penalized for making the same mistake multiple times in a word.
Simple averaging (adding together individual character accuracy values 250 and dividing by the total number of characters) of the character accuracy values 250 may also be used to calculate word grouping accuracy values 260. However, simple averaging may not provide an accurate representation of the OCR's recognition. In the above example. the word grouping accuracy value 260 of "ear" would be .85, and the value of "earring" would be .84 using simple averaging. The user 180 would believe that "ear"
was recognized with greater accuracv than "earring." However, in light of the above discussion. this belief is erroneous. This is explained by observing that an inaccurately recognized character is more detrimental to the word recognition of shorter words than longer words. Thus. if one of three letters is misrecognized, the word may be completely misinterpreted. However. if one of seven letters is misrecognized. the word may still be recounized correctlv based on its surrounding letters, using a fuzzy matching system.
Altemate methods of calculating word grouping accuracy values 260 may be beneticially employed depending upon the user 180 and the system. Figure 5c illustrates I 9100-3 I 7T/709 I 13.1 a second method of calculating word grouping accuracy values 260. A word grouping is selected 540. A character accuracy value 250 is obtained 544 for a first character in the selected word grouping. The system determines 548 whether the obtained character accuracy value 250 is greater than a threshold accuracy level. The threshold is set by the user 180 to define an overall accuracy level for the word accuracy generation system. If the obtained character accuracy value 250 is greater than the threshold. a counter is incremented 552. If the obtained value 250 is less than the threshold, the counter is not incremented. In either case, the system determines 556 whether this was the last character in the word grouping. If there are more characters. the system increments 560 a counter and repeats the process for the next character.
If there are no more characters, the system determines 564 whether all of the characters in the selected word grouping passed the threshold. If all of the characters did pass the threshold. the system calculates 568 the word grouping accuracv value 260 of the selected word grouping as the average of the character accuracy values 250. If at least one of the characters did not exceed the threshold. the system determines 572 the percentage of characters that did not exceed the threshold, using the formula (P-N)/100, where P is the total number of characters in the word grouping, and N is the number of characters which exceeded the threshold. Then the system calculates the word grouping accuracy value 260 of the selected word grouping by subtracting 576 the result from the threshold accuracy value. This process is repeated for each word grouping.
An advantage to using this method over assigning binarv representation is that this method allows word grouping accuracy values 260 to be searched using any integer.
19100-3177//7U91I3.1 In the tirst method tor calculatini-, ~tord grouping accuracy -,alues 260. the values are quantized or digitized. Thus, ita user 180 %~anted to search ~tiords created having a range of values from .34 to .4-1. the tirst niethod is unlikelv to retrieve anv word groupings. as word grouping accuracy values 260 created using its scheme are the result of some fraction such as 3/4, or 6/7. 1 lowever. the values 260 stenerated by the second method are calculated as the average of individual character accuracy values 250, which may be any integer over the threshold, or values 260 are calculated as the result of the percentage subtracted from the threshold value. v~hich also generates a broader range of results.
Thus a user 180 searching for a range ot values is more likelv to retrieve results in a system using the second method rather than the tirst method.
Figure 5d displays a third method of calculating word grouping accuracy values 260. A word grouping is selected 580. A character is selected 581 from the selected word grouping. The accuracv value 250 of the selected character is obtained 582. The system then determines 584 %ti-hether the accuracy value 250 of the selected character is greater than a first threshold. If it is, the character is assiened 590 a"*".
Any symbol may be assigned to the character. However, as the tirst threshold is the highest threshold, the symbol should visually represent the fact that the character is in the most accurate category. After assigning 590 the "*". the svstem repeats 598 for the next character. or if it was the last character. the system repeats 599 tor the next word grouping.
If it ~vas the last character in the last word grouping, the svstem proceeds to the display step ? 16.
lf the sclected character did not exceed the first threshold. the svstem determines 586 hether the character's accuracy value 250 exceeds the second tllreshold. I
t it does, then a "+" is assigned 592. Again, any symbol may be used but the symbol should represent to the user 180 that the character is in the second best category.
If the character does not pass the second threshold, the system compares 588 the character to a third threshold. If the character exceeds the third threshold. a"-" is assigned. If the character fails to exceed the third threshold, a "-" is assigned. After all of the characters have been assigned a symbol. the words and their symbols are displayed 216 to the user 180.
Word grouping accuracy values 260 generated in accordance with this method are advantageous because they visually communicate to the user 180 the quality of the characters in a given word. Thus, if a user 180 sees a word and the symbols "**+*"
displayed. the user 180 knows that the word has been recognized accurately.
This visual representation may be more effective in communicating accuracy values 260 than numeric displays. The thresholds in the above method are preferably set by the user 180 to suit the user's preferences. One of ordinary skill in the art can see that other methods ot'calculating word grouping accuracy values 260 may be used in accordance with this 1~ invention. For example, symbols may be assigned to words instead of characters, allowing the user 180 to have a visual representation of word grouping accuracy values 260. Or. the word grouping accuracy value 260 of a word may be set equal to the accuracy value 250 of the character in the word grouping which has the lowest accuracy %,alue 250.
~0 For all of the above methods, after the word grouping accuracy value 260 calculated tor a word grouping, the value and its associated word grouping are stored in a 19100-3177//709113.I
table 170 located on disk 160. The table 170 is used to retrieve the accuracy value 260 for a word grouping when the value 260 needs to be adjusted or displayed.
Figure 6a describes one embodiment of the adjusting word accuracy module 416.
In this embodiment. the word grouping accuracy value 260 of a selected word grouping is adjusted based upon comparisons of the selected word grouping to word groupings on a stored list. The stored list is preferably composed of commonly used words such as those found in a dictionary. Additionally, words or acronyms that are commonly used in the tvpes of documents beine processed by the system are preferably added to the stored list by the user 180. For example, if a company using the present invention is a software development companv. thev mav create a list including words such as "debug", "beta", and "GUI." The additional or custom words are added to the dictionary word list, or. in a preferred embodiment. the custom words are stored in a separate word list.
In accordance with the present invention, a first word grouping is selected 600.
The word grouping is compared 604 to a word grouping in the stored list. The system determines 608 whether there is a match between the selected word grouping and the word grouping on the stored list. If there is a match. a constant is added 628 to the accuracy value 260 of the selected word grouping. The system determines 632 if there is another word grouping to be compared, and if there is. a counter for the selected word groupings is increased 636. the counter for the stored list is set 640 to one, and the counter for the stored list is set 644 to one. If the system determines that there are no more word groupings to be compared, the system moves to the display module 216.
19100-3177.'/7091t3.1 If there xvas not a match between the selected word grouping and the word grouping of the stored list, the system detertnines 612 if this the last word grouping on the stored list. If there are more word groupings on the stored list, a counter is incremented 616. and the next word on the list is compared with the selected word grouping. If there are no more word groupings on the stored list, the system determines 620 if there are more stored lists. If there are more stored lists, a counter is incremented 622 and the comparison process is repeated for the new stored list. If there are no more stored lists, no value is added to the selected word grouping, and the system moves on to the next word grouping in process step 632.
In the example of the word "earring," the word grouping accuracy value 260 was 0.71. If the above module 416 were applied, "earring" would be compared to a stored list of words. I f a dictionary was one of the stored lists, a match would be found. The adjusting module 416 would increase the value of the word accuracy to .81 because of the match, using a constant of .05. The user 180 would then see the increased confidence factor, and feel more confident that the word was accurately recognized.
However, if in the original example. the word was "GUI", and the "I" was misrecognized as a"L", the word "GUL" would have been created. The OCR confidence factor for the "L" may still be high, for example, at .75, because the OCR may believe that the "L" was recognized correctly. A user 180 who sees "GUL" who is not the author of the document may be unsure w-hether GUL is a new acronym or a misrecognized word. However. a stored list comparison containing technical words and acronyms would not find a match for "GUL", and therefore. the value would not be adjusted. Since words correctly recognized will 1 9 1 00-3 1 77 ,'7091 13.1 have a .5 value added. the %alue for "GUL" %%ill appear lo%~ to the user 180. -I'he user 180 %%ill have less contidence that the word "GUL" is accuratelv recoenized.
and mav choose to exclude "GUL" from beine used to index the document. Thus, one advantaLye of this system is that it allo~ks non-experts to perform document entry. In another embodiment of the system, a constant is subtracted from the %tiord accuracy measure if no match is found. In either embodiment. the system is optimized if the authors of documents update the word lists trequently with the acronvms and special word uses of the authors.
In Figure 6b, an alternate embodiment of the adjusting -word accuracy module is illustrated. In this embodiment. language rules are applied to the word grouping to adjust the word grouping accuracy value 260 of the selected word grouping. Language rules are rules such as no capital letter in the middle of a word. For example. if an OCR
recognizes the word "example" as "examPle". the application of a language rule would catch the mistake and subtract a constant from the word grouping accuracy value 260.
Other language rules mav include "no i before e except atter c." the first letter of' the tirst word of a sentence should be capitalized, etc. The list of language rules mav be edited by the user 180 to create language rules that may be specific to the user's field. For example, if the present invention is used for a foreign language. foreign language rules may be implemented.
As illustrated in Figure 6b. a kvord grouping is selected 648. 'I'he system determines 652 if the word grouping contradicts the tirst language rule. If'it does. a value is subtracted 656 froni the accuracv value 260 of the sc:lected word grouping.
I f the %~ord grouping does not contradict the first language rule, the system determines 654 if there are more language rules. If there are, a counter is incremented 668, and the next language rule is applied. If there are no more language rules to be applied or a language rule was applied, the system determines 660 if there are more word groupings. If there are not, the process proceeds to the display module 216. The above embodiment may be used alone or in conjunction with the stored list comparisons to provide a more complete word accuracy generation system. Other types of word accuracy checks may be used in accordance with the present invention.
Display module 216 displays the recognized word groupings and their associated word grouping accuracy values 260 to the user 180. This gives the user 180 an opportunity to understand how accurately the OCR recognized the document. If the OCR
recognized the words poorly, the user 180 may be forced to delete the OCR
version, and use the paper document in its paper form. Alternatively, the user 180 may employ another OCR to obtain a better result. If the OCR repeatedly generates inaccurate word ~~roupings. the user 180 may decide to purchase another OCR system.
Keeping track of the OCR's performance is also important in an embodiment wherein the recognized word groupings are used to index a document. In this embodiment, an indexing list of the recognized word groupings is used to retrieve documents. The word groupings on the list are used as potential search terms for 10 retric%-ing the document. For example, if the word "rock" was in a word list for I)ocumcnt A. the word "rock" may be entered as a search term to find and retrieve 19100-3177//709113.1 Document A. The indexing list is formed from the table 170 of word groupings and their accuracy values 260.
To begin the process. the table 170 is displayed to the user 180. The user 180 can then determine whether or not to remove some of the word groupings from the table 170.
Words removed from the table 170 will not be on the indexing list. For example. the user 180 may want to prevent any word that does not appear to the user 180 to be an accuratelv recognized word from the list so it cannot be used to retrieve the document.
This ensures that the word list used to index the document is kept free of "pollutants,"
which are words that are misrecognized and may introduce error into the file retrieval process. For example. if the word "ear" is on the list for Document A. but the true word was "eat", entering "ear" as a search term will retrieve Document A
erroneously.
Ho-wever. if "ear" is displayed with a low word grouping accuracy value 260, the user 180 may choose to remove that word based upon its accuracy value 260 from the word list for Document A.
In the preferred embodiment, the user 180 simply specifies a level of accuracy value 260 that the user 180 would like to maintain in the user's word lists.
For example, the user 180 may specify that all words that are recognized at less than .95 should not be kept. In this embodiment, after a word grouping has its final accuracy value determined. its accuracy value 260 is compared to the threshold. If it passes, the word is kept in the indexing list. If it fails, it is not stored in the indexing list.
Then. the user 180 does not have to view the results of the word accuracy generation. and does not have to perform any subsequent nianual determinations. Again, this introduces an ease uf use to 19100-3177~ ~ 7091 13. I
the document tile indexing system. as a user 180 without expertise mav be used as the document imaging/filing clerk.
Another embodiment of the present invention displays the image of the document to the user 180 using color codes to indicate the accuracy values 260 of the words. For example. one color, such as red, is assigned to words that have accuracy values 260 over a threshold of 0.5. and blue may be assigned to words having an accuracy value 260 of less than 0.5. This allows the user 180 to view the image of the document and visually understand which words are correctly recognized and which are not. The user 180 may set the threshold to whatever level of accuracy the user 180 requires.
Figure 7a illustrates an embodiment of the present invention wherein multiple optical character recognition techniques are employed. In this embodiment. the present invention calculates word grouping accuracy values 260 of each word grouping created by each OCR . These word grouping accuracy values 260 can then be used to determine the most accurate version of the word grouping.
The system begins by receiving 700 data, and creating a bitmapped image of the paper document. An OCR program is applied 704 to the bitmapped image to recognize the characters in the image. This is repeated for each OCR present in the system. Thus, multiple versions of the original document are created.
The system then obtains 708 character accuracy values 250 for each character in the first version of the original document. Next, word groupings are created 712 f'rom the first version of the document. Initial word grouping accuracy values 260 are calculated 716 for each created word grouping. The initial word grouping accuracy values 260 are 1 9 1 00-3 1 77//709 1 1 3 .1 calculated in accordance with the process described above in relation to Figures 5a-d.
The %%ord groupings and their accuracv %alues 260 are stored in a table 170.
The process is repeated 720. 722 for each OCR version. Each OCR version has its okti-n table 170 composed of word groupings and accuracy values 260.
The svstem then compares the different versions of each ~.vord grouping together to create tables 850 of versions of the word groupings ordered bv accuracv.
After an initial ordering has been created, the accuracy values 260 for each version of the v,-ord grouping are adjusted 728 in accordance %vith predefined accuracv factors.
Then. the table 850 is reordered based on the new accuracy values 260.
In one embodiment, composite word groupings are created 732. Composite word groupings are word groupings created from the most accurate versions of each character in the word grouping. Thus. tor example, if the word "help" %vas recognized by the present invention, each OCR would generate character accuracy values 250 for each of the four characters. OCR A mav generate "h": .87 "e": .92 "1": .95 "p": .77 and OCR B
mav eenerate "h": .91 "e": .90 "I": .90 "p": .82. Therefore. if OCR A and B-,kere the onlv OCRs present in the system, a composite ord grouping is created using the "h"
from OCR B. the "e" from OCR A. the "I" from OCR A. and the "p" from OCR B.
Finally, the most accurate word groupings 736 are displayed or stored into an indexing list. If composite %tiord groupings are used. a word grouping accuracv value 260 is calcu(ated for the composite word grouping and is compared with the other ,kord groupings to determine the most accurate word groupings.
, ~
_~
19 1 00-3 1 77." ',091 13 .1 Figure 8 illustrates ordering the versions of the word groupings in greater detail.
A word grouping is selected 804 from the table 170 of word groupings created by the first OCR. The coordinates of the selected word grouping on the bitmapped image are determined 808. This process is performed in order to ensure that versions of the same ~ word grouping are compared against each other. The version of the word grouping is then stored 812 in a first word grouping version table 850. A counter is incremented 816, and a next table 170 of word groupings created by a next OCR is selected 820.
A word grouping is selected 824 that corresponds to the coordinates determined previously. The accuracv value 260 of the selected version is compared 828 to the accuracy value 260 of the version of the word grouping already stored in the word grouping version table 850.
The system then determines 832 whether the selected version has a greater accuracy value 260 than the version in the table 850. If it has, the selected version is stored 844 in the table 850 in a position immediately prior to the first version. If the selected version has a lesser accuracy value 260, the system determines 836 if there are more versions of the word grouping stored in the word grouping version table 850. If there are not, the selected version is stored 848 in a position immediately after the first x=ersion. If there are multiple versions of the word grouping already stored in the word grouping version table 850, the selected version is compared against all of the versions in the table 850 until a version is found that has a greater accuracy value 260.
At that point, '0 the ~. ersion is stored in the table 850 in a position after the version having a greater accuracy value 260. For example, as illustrated in Figure 8b, if the word "example" was in the original paper document, and there were OCRs A. B. C, and D present in the 19100-3177//7091 13 .1 system, there would be four versions of the word "example" created. The system creates a table 850 of versions of the word "example." and orders the versions by accuracv.
Thus. as shown. the calculated word grouping accuracy values 260 for the versions of example are: OCR A"example": .91. OCR B "example": .87. OCR C "example": .72 and OCR D "example": .95. The table 850 therefore orders the versions beginning with OCR
D's "example". followed bv A's, B's. and C's. If a fifth version of "example"
is created by a fifth OCR. the fifth version would be compared to the four versions already in the table 850. The accuracy value of the fifth version is compared to the accuracy values of the other four. and is placed into the table accordingly. If the fifth version has an accuracy value 260 of.89. its version is placed in a position under OCR A's version and above OCR D's version.
The system then determines 852 if there are more OCR versions of the selected -ord Llrouping. [f there are. a next version of the word grouping is selected, and the process is repeated. If there are no more versions of the word grouping, the system determines if there are more word groupings in the first OCR word grouping table 170. If there are. the next word grouping is selected, and the process is repeated.
Figure 9 illustrates adjusting accuracy values 260 after the versions of each word grouping have been ordered. First, a factor or constant is added 900 to a version of a word grouping based upon the strengths of the OCR which created the version of the word grouping. For example. if the word "help" has been created by OCR A. and OCR
A recoenizes Courier font particularly well, the word grouping accuracy value 260 for the version of help created by OCR A is adjusted upwards by a constant. This is repeated for '5 19100-3177,'/ 7091 13.1 each version of the word grouping in the word grouping table 170. Then the versions of the word grouping on the word grouping table 170 are reordered 904 based upon the changes in accuracy values 260.
A factor or constant is subtracted 908 from the versions based upon OCR weaknesses in the manner described above. The table 170 is reordered 912 based upon any change in accuracy values 260. Then, the word grouping accuracy values 260 are adjusted and reordered based upon stored list comparisons 916 and language rule applications 920, as described in Figures 6a and 6b. The adjusting and reordering produces a final ordering of word grouping versions from most accurate to least accurate. At this point, all word groupings having an accuracy value over a threshold may be kept, or only the most .15 accurate word grouping version may be kept.
Figure 10 illustrates the display word grouping aspect of the present invention, which may be used in a single or multiple OCR embodiments of the present invention. However, it is described here with respect to the multiple OCR
embodiment of the present invention. A first word grouping version table 850 is selected 1000. The version in the first position in the table 850 is retrieved as it is the most accurate version based upon the ordering performed earlier.
In one embodiment, the composite version of the word grouping is compared to the retrieved version to determine which is the most accurate version. This embodiment may be made into an optional step dependent upon the accuracy of the most accurate version of the word grouping. If the most accurate version of the word grouping has an accuracy value 260 greater than a threshold, no composite word grouping is created. This saves on processing time, as composite word grouping creating may be a time-consuming process for low-speed processors. The threshold may be specified by the system, or is preferably a user-defined threshold level.
Finally, the most accurate version of the word grouping is displayed. Alternatively, the most accurate version is used to create a list used for indexing the document, as shown in Figure 11. In this embodiment, if the system determines 1100 that a version of the word grouping has an accuracy value which exceeds a threshold, the version is used to update 1104 an indexing list. Again, the threshold is a level of accuracy the user 180 desires for the user's system. By keeping all of the versions which exceed the threshold, the user 180 can eliminate, to the greatest extent possible, OCR mistakes which are not caught by the above-described processes. For example, if the table 850 contains three versions of the word "ear"
which exceed the threshold, and the most accurate version is "eat", and the other two are "ear", a system which keeps the version with the highest accuracy value will incorrectly keep "eat" as the version of the word. A system which keeps all three will also keep the versions which have the correct representation of the word "ear." Thus, in those systems, the document is able to be retrieved by using "ear" as a search term. However, by keeping multiple versions, more disk space is used in keeping the longer indexing lists.
If none of the word groupings exceed the threshold, in either the single or multiple OCR embodiments of the present invention, the user 180 is preferably displayed 1108 several options regarding how to proceed. After receiving 1112 an input, the system executes the option selected. A first option 1116 is to perform the OCR again, in an attempt to increase the character accuracy values. If multiple OCRs are present in the svstem, the user 180 may select one of the other OCRs. A second option 1120 is to re-scan the document. This may provide a cleaner image for the OCR to recognize.
A third option 1124 is to allow the user 180 to lower the threshold. If the word groupings consistentlv fail to meet the threshold, the user 180 may have set the threshold too high for the user's svstem.
The above description is included to illustrate the operation of the preferred embodiments and is not meant to limit the scope of the invention. The scope of the invention is to be limited only by the following claims. From the above discussion, many variations will be apparent to one skilled in the art that would yet be encompassed by the spirit and scope of the present invention.
What is claimed is:
19100-3177//7091 13.1
Background Art The field ot'document imaging is growing rapidiv, as modern society becomes more and more digital. Documents are stored in digital format on databases, providing instantaneous access, minimal phvsical storage space, and secure storage.
Todav's society now faces questions on how best to transfer its paper documents into the digital medium.
The most popular method of digitizing paper documents involves using a system comprising a scanner and a computer. The paper documents are fed into a scanner. which 1 -S creates a bitmap image of the paper document. This bitmap image is then stored in the computer. The computer can take a varietv of forms. including a single personal computer (PC) or a network of computers using a central storage device. The bitmapped images must be able to be retrieved after they are stored. One system for filing and retrieving documents provides a user interface which allows a user to tvpe in a search 10 term to retrieve documents containing the search term. Preferablv, the svstem allows the user to type in any word that the user remembers is contained within the desired document to retrieve the desired document. 1{owever, in order to retrieve docunients on this basis. the document must be character recognized. That is. the computer must recognize characters within the bitmapped image created by the scanner.
Another common usage of digitizing documents is to digitize long paper documents in order to allow the document to be text searched by the computer.
In this usage, a user types in the key word the user is looking for within the document, and the system must match the search term with words found within the document. For these systems. the document must be character recognized as well.
The most common method of recognizing characters is by using an optical character recognition (OCR) technique. An optical character recognition technique extracts character information from the bitmapped image. There are many different types of optical character recognition techniques. Each has its own strengths and weaknesses.
For example. OCR I may recognize handwriting particularly accurately. OCR 2 may recognize the Courier font well. If OCR I is used to recognize a document in Courier font, it mav still recognize the majority of the characters in the document.
However. it may recognize many of the characters inaccurately. A user may not know of an OCR's strengths and weaknesses. A user may not know whether or not the types of documents the user typicallv generates are of the kind that are accurately recognized by the OCR
present on the user's system. Current systems do not inform the user of the quality of the recognition of the OCR technique. The user finds out how accurate the recognition was only by using the document for the purpose for which it was stored into the computer system, at which time it may be too late to correct.
I 9100-3177/ /7091 13.1 An inaccuratelv recounized document can lead to several problems. First of all, in a system in which documents are stored and retrieved based on their contents.
an inaccurately recognized document may become impossible to retrieve. For example. if a user believes the word "imaging" is in a specitic document. the user will type in "imaging" as the search term. However, if the word "imaging" is recognized incorrectly.
such that it was recognized as "emerging." the user's search %vill not retrieve the desired document. The user may not remember anv other words in the document. and thus the document is unretrievable. In a svstem xhere documents are dittitized to allow text searching of the document, the same problem occurs. Misrecognized words are not tound by the use of the correct search terms.
Thus, there is a need to allow the user to determine whether a recognized word is of acceptable quality. By allowing the user to determine whether akvord is of acceptable quality. the user can ensure that the document is retrieved by the use of that %vord as a search term. Also. a user can ensure that words within the document are accurately recognized for internal document searching purposes. Additionallv. in a system . ith multiple optical character recognition techniques. there is a need to be able to compare the accuracy of the different versions of the document to create a version that is the most accurate.
Disclosure of the Invention The present invention is a computer-implemented method lor calculating \~ord groupinu accuracy values (260). l'he present invention receives (200) data.
pertorms (204) an optical character recoi-nition technique upon the received data. and creates (208) I9I00-3177 'O9113.1 word groupings. The system then calculates (212) word grouping accuracy values (260) for the created %vord groupings.
Word grouping accuracy values (260) are calculated (212) by using character accuracy values (250) determined by the OCR technique. The present invention preterably uses these character accuracy values (250) to create a word grouping accuracy value (260). Various methods are employed to calculate the word accuracy (260), including binarizing the character accuracy values (250), modified averaging of the character accuracy values (250), and employing fuzzy visual displays of word grouping accuracv values (260). The calculated word grouping accuracy values (260) are adjusted based upon known OCR strengths and weaknesses, and based upon comparisons to stored word lists and the application of language rules. Word grouping accuracy values (260) are normalized and displayed or compared to a threshold. The words whose accuracy values (260) exceed the threshold may then be used to index the documents or provide search terms tor searching within the document. If no word groupings exceed the threshold then the user is offered different options, including to clean the image by performing another OCR or scanning the document again, or to reset the threshold to a lower value.
Brief Description of the Drawinps -)0 These and other more detailed and specific objects and features of the present invention are more fully disclosed in the following specification, reference being had to the accompanying drawings. in which:
19100-3177//709113.1 Figure 1 is a block diagram of a computer system embodying the present invention.
Figure 2a is a flowchart diagram illustrating the process steps of an embodiment of the present invention in which word accuracy is calculated for a single OCR
system.
~ Figure 2b is an illustration of two characters and their accuracy values 250.
Figure 2c is an illustration of a word grouping and its individual character accuracy values 250.
Figure 3 is a more detailed flowchart of process step 208 of the present invention.
Figure 4 is a more detailed flowchart of process step 212 of the present invention.
Figure 5a is a more detailed flowchart illustrating the assigning binary representation module 412 of the present invention.
Figure 5b is an illustration of word groupings and accuracy values 260 in accordance with the present invention.
Figure 5c is a flowchart of one alternate embodiment of calculating word grouping accuracy values 260 in accordance with the present invention.
Figure 5d is a flowchart of a second alternate embodiment of calculating word grouping accuracy values 260 in accordance with the present invention.
Figure 6a is one embodiment of the adjusting word accuracy module 416 of the present invention.
Figure 6b is an alternate embodiment of the adjusting word accuracy module 416 of the present invention.
19100-3177'1709113.1 Figure 7 is a flowchart diagram illustrating the process steps of an embodiment of the present invention in which word accuracy is calculated for a multiple OCR
system.
Fieure 8a is a more detailed flowchart of process step 724 of the present invention.
Figure 8b is an illustration of a word grouping version table 850.
Figure 9 is a more detailed flowchart of process step 728 of the present invention.
Figure 10 is a more detailed flowchart of process step 736 of the present invention.
Figure 1 I is a flowchart illustrating an updating indexing list embodiment in accordance with the present invention.
Detailed Description of the Preferred Embodiments Figure 1 illustrates a hardware embodiment of the present invention. A data receiver 100 is coupled to the data bus 110 of a computer 120. The data receiver 100 preferably receives data from a scanner 104, which can be any conventional scanner, such as Canon DR 3020 or Canon PICS 4400. Alternatively, the data receiver can be a network adapter or [nternet connection comprising a modem and a phone line which couples the computer 120 to a remote data source from which data is received.
The computer 120 includes a central processing unit 130, such as an Intel Pentium processor, a displav 140. an input device 190, random access memorv (RAM) 150, read-onlv memory (ROM) 155, and a disk 160, all interconnected through bus 1 10.
1 9 1 00-3 1 77 .'7091 13.1 txn input device 190 is a mouse. keyboard. or other pointine device that allows a user 180 to interact Nk-ith the coniputer 120. ROM 155 provides CPU 130 -with unvarying tunctions such as executine programs initialized at the start-up otcomputer 120. R.AM
150 provides the CPU 130 with data and instruction sequences that are loaded trom disk 160. The disk 160 is. for example, 1.6 Gigabyte disk that intertaces to the computer bus throui-yh a Small Computer Svstem Interface (SCSI) interface 165. Disk 160 stores data tiles such as binarv. grav-scale. and color image tiles, a table 170. text tiles, and proerams includine a word accuracv calculation program and one or more OCR programs.
Disk 160 is a single device or multiple devices. The components of computer 120 take conventional form. Computer 120 can stand alone or is a part of a network of computers 125 comprised of individual computers 120 coupled together through Ethernet connections. The present invention can be used in conjunction with any computer operating system. such as Windows 95. Windows NT. or OS/2, resident in memorv 155.
As illustrated in Figure 2. the present invention receives 200 data. As described above, data can 200 be received in a multitude ot wavs. including by a scanner or hv a remote connection with another computer. I f a scanner 104 is coupled to data recei% er 100, then the paper document fed into the scanner 100 is digitized. Digitizing a paper document involves creating a bitmap of the paper document. Atter a digitized version of the document is created, an OCR is pertormed 204 upon the bitmap. A Canon OCR
technique. specit7ed in U.S. 11'at. 5.379,349, is preferably used as the OCR
technique:
however. any OCR technique can he used in accordance %vith the present invention. The OCR technique creates character information trom the bitmap. :\ description othow the character recognition process is accomplished is found in the above-mentioned U.S. Pat. 5,379,349. A document may be a single piece of paper or multiple pieces of paper that are substantially related.
Most OCRs provide a character accuracy value 250 for characters recognized from the bitmap. The character accuracy value 250 is typically created by comparing the recognized character with a stored template of the character.
The differences in the two versions of the character are quantized and represented as an accuracy value 250. For example, a recognized "p" and a stored template of a "p" are shown in Figure 2b. This "p" is given a character accuracy value 250 of .95 because of the differences in shape between the recognized "p" and the stored template of the "p".
The system next creates 208 word groupings from the recognized characters, as illustrated by process steps 300, 304, 308, 312 and 316 shown in Figure 3.
After word groupings are created 208 from the extracted character information of the bitmap of the paper document, the system calculates 212 the word grouping accuracy values 260 of the created word groupings. Word grouping accuracy values 260 are calculated to provide the user 180 with an easily understood measure of the quality of an OCR's recognition. Displaying a string of character accuracy values 250 may not communicate useful information to the average user 180. For example, in Figure 2c, the word "rock"
is composed of four characters "r", "o", "c", and "k." The "r" is recognized at a .95 accuracy value 250, the "o" is recognized at a .72 accuracy value 250, the "c" is recognized at a .92 character accuracy value 250 and the "k" is recognized at a .82 accuracy value 250.
Displaying individual character accuracy values 250 does not inform the user whether the Ntiord "rock" %tas accurately recognized. The user 180 knows that "r" and "c"
were accuratelv recognized. but the user 180 does not know how the poor recognition of the "o" and the "k" affects the recognition of the word as a whole.
Determining the accuracy otthe word, as opposed to individual characters, is critical in systems that allow retrieval of documents bv the use of text-searching. In those systems. when the user 180 searches tor a document, the user 180 enters a search tetm composed of words.
Thus, it is more important to know whether an entire word is accurately recognized. as opposed to whether individual characters are accurately recognized. Additionally, in documents in which OCR techniques are performed in order to allow the document itself to be text searched. it is important to know whether the OCR has accurately recognized the document. If the OCR recounition is inaccurate, words searched for will not be tound.
even if the words are present within the image. Finally, if multiple words are displaved with all of their individual character accuracy values 250, the page soon becomes an indecipherable conglomeration of letters and numbers.
Figure 4 illustrates the process steps of an embodiment of the current invention calculating word accuracy in a system employing a single OCR. The svstem selects one ot the newlv created word grouping using a selecting module 400. The svstem then obtains the character accuracv value 250 for one of the characters in the selected word grouping by using an obtaining character accuracv value module. As described above.
1 9100-3 1 77-,'7091 13.1 character accuracy values 250 are typically available after perforrning an OCR. The system then applies a character accuracy value adjusting module 408 to adjust the character accuracy value 250 based upon known OCR weaknesses. This adjustment is based upon the fact that OCRs typically have unique strengths and weaknesses.
The system determines whether the character being examined is one of the known characters that is not accuratelv recognized, or is a font, typeface, or style that is not accurately recognized. In this situation, the system subtracts a constant from the obtained character accuracv value 250. The constant is typically .05, but can be set higher or lower, depending upon user preferences. For OCR strengths, the adjusting module 408 adjusts the character accuracy value 250 upwards to reflect characters, fonts, or typefaces that the OCR typically recognizes more accurately than other recognition systems In the example of Figure 2c, perhaps the word "rock" was created using the Times New Roman font, and the OCR applied to the document recognizes the Times New Roman font accurately. Then. the individual character accuracy values 250 are adjusted upwards to become "r": 1.0, "o": .77. "c": .97, and "k": .87. Thus, although the OCR had a confidence factor of only .72 that the "o" was really an "o". the system of the present invention raises the value to .77 due to the strength of that OCR's ability to recognize Times New Roman. If this number were displayed to the user 180, the user 180 would feel more contident that the "o" was correctly recognized because of the adjustment of the present invention.
Of course. if a document contains words created in one font that is either strongly or weaklv recognized by the OCR present in the system, than all of the characters are 19100-3 I 77,'/7091 13 .1 adjusted uniformlv. and the adjustment is factored out. tlowever. in multiple OCR
situations, discussed belok~. it is important that the adjustment occur. as the different versions of the word groupings created by each OCR are compared against each other.
The above process is repeated tor each character in the selected -ord grouping. and is ~ repeated for each word erouping in the docunlent.
An assigning module 412 is then applied to assign a binary representation to the characters in the selected word grouping. discussed in more detail below. A
preliminary word strouping accuracy value 260 is therebv generated for the selected word grouping.
A word accuracy adjusting module 416 is applied to the selected \,vord grouping to adjust the preliminary word grouping accuracy value 260 based upon predefined tactors such as stored list comparisons or the application of stored language rules. discussed in more detail below. This process is preferably repeated for each word grouping in the received group of data. After a word grouping accuracy value 260 is calculated for the last v.-ord grouping. the system proceeds to displav the word groupings and their accuracv values 260 using a displav module 216. The above-described modules mav be implemented as hard-ware. firmware. or soft%vare, in accordance with the present invention.
Figure 5a illustrates in more detail the assigning binary representation module 412. The svstem selects 500 a character in the selected word grouping. Then.
the system determines 504 whether the tirst character in the word grouping has an accuracy value 1-50 that exceeds a threshold accuracy level. The threshold accuracy level is preferably user detined. The user 180 can theretore set the recognition quality ot the entire s~ stem based upon what threshold the user 180 chooses. For example. it the user 180 desires a I 9101)- 3 177 709113.1 svstem that accepts characters at 90% accuracy, the user 180 can set the threshold at .9.
I f the user 180 is more tolerant of error, the user 180 can set the threshold lower.
The svstem compares the accuracy of the selected character to the user-defined threshold to determine what representation to assign the character. If the character ~ accuracy value 250 exceeds the threshold, the character is assigned 508 a"1." If the character accuracy value 250 does not exceed the threshold, the character is assigned 512 a"0." The system determines 516 if there are more characters. If there are, a counter is incremented 520. and the process is repeated. If all of the characters have been assigned a representation. the system then adds 524 the number of characters assigned a"one"
together to determine the total number of characters assigned a "one". Then the system determines 528 the total number of characters in the word grouping. Finally, the system calculates the preliminary word accuracy for the word grouping by dividing 532 the number of characters assigned a "one" by the total number of characters in the word grouping.
1~ The system uses the above method of calculating because it normalizes word grouping accuracv values 260. Normalizing provides a better representation of the recognition quality of a word. For example, in Figure 5b, the word "ear" has character accuracy values 250 of.93, .89, and .72 assigned to its characters. The threshold is .8.
Theretbre. the "e" and the "a" are assigned a 1 and the "r" is assigned a 0.
The word grouping, accuracy value 260 for "ear" is therefore .67 (2/3). The next word is "earring."
I'he charactcr accuracy values 250 are the same for the first four letters, the "i" is recognized at a.9;. the "n" is recognized at a .85. and the "g" is recognized at a .87.
1 9 1 00-3 1 77//709 1 1 3. t Therefore, its word accuracv is .71 (5/7). The percentage of characters with an accuracy ~ alue above the threshold is greater in the word "earring" than in the word "ear." This is retlected in the -,%ord accuracy value assigned to each word. In one embodiment, repeated characters are discounted for the purposes of calculating word grouping accuracy values ~ 1-60. 1 n the above example, the "r" would only be counted once in calculating the accuracy value, leading to an accuracy value of .83 (5/6). This embodiment is useful when the user 180 is attempting to evaluate the accuracy of the OCR, as the OCR is not penalized for making the same mistake multiple times in a word.
Simple averaging (adding together individual character accuracy values 250 and dividing by the total number of characters) of the character accuracy values 250 may also be used to calculate word grouping accuracy values 260. However, simple averaging may not provide an accurate representation of the OCR's recognition. In the above example. the word grouping accuracy value 260 of "ear" would be .85, and the value of "earring" would be .84 using simple averaging. The user 180 would believe that "ear"
was recognized with greater accuracv than "earring." However, in light of the above discussion. this belief is erroneous. This is explained by observing that an inaccurately recognized character is more detrimental to the word recognition of shorter words than longer words. Thus. if one of three letters is misrecognized, the word may be completely misinterpreted. However. if one of seven letters is misrecognized. the word may still be recounized correctlv based on its surrounding letters, using a fuzzy matching system.
Altemate methods of calculating word grouping accuracy values 260 may be beneticially employed depending upon the user 180 and the system. Figure 5c illustrates I 9100-3 I 7T/709 I 13.1 a second method of calculating word grouping accuracy values 260. A word grouping is selected 540. A character accuracy value 250 is obtained 544 for a first character in the selected word grouping. The system determines 548 whether the obtained character accuracy value 250 is greater than a threshold accuracy level. The threshold is set by the user 180 to define an overall accuracy level for the word accuracy generation system. If the obtained character accuracy value 250 is greater than the threshold. a counter is incremented 552. If the obtained value 250 is less than the threshold, the counter is not incremented. In either case, the system determines 556 whether this was the last character in the word grouping. If there are more characters. the system increments 560 a counter and repeats the process for the next character.
If there are no more characters, the system determines 564 whether all of the characters in the selected word grouping passed the threshold. If all of the characters did pass the threshold. the system calculates 568 the word grouping accuracv value 260 of the selected word grouping as the average of the character accuracy values 250. If at least one of the characters did not exceed the threshold. the system determines 572 the percentage of characters that did not exceed the threshold, using the formula (P-N)/100, where P is the total number of characters in the word grouping, and N is the number of characters which exceeded the threshold. Then the system calculates the word grouping accuracy value 260 of the selected word grouping by subtracting 576 the result from the threshold accuracy value. This process is repeated for each word grouping.
An advantage to using this method over assigning binarv representation is that this method allows word grouping accuracy values 260 to be searched using any integer.
19100-3177//7U91I3.1 In the tirst method tor calculatini-, ~tord grouping accuracy -,alues 260. the values are quantized or digitized. Thus, ita user 180 %~anted to search ~tiords created having a range of values from .34 to .4-1. the tirst niethod is unlikelv to retrieve anv word groupings. as word grouping accuracy values 260 created using its scheme are the result of some fraction such as 3/4, or 6/7. 1 lowever. the values 260 stenerated by the second method are calculated as the average of individual character accuracy values 250, which may be any integer over the threshold, or values 260 are calculated as the result of the percentage subtracted from the threshold value. v~hich also generates a broader range of results.
Thus a user 180 searching for a range ot values is more likelv to retrieve results in a system using the second method rather than the tirst method.
Figure 5d displays a third method of calculating word grouping accuracy values 260. A word grouping is selected 580. A character is selected 581 from the selected word grouping. The accuracv value 250 of the selected character is obtained 582. The system then determines 584 %ti-hether the accuracy value 250 of the selected character is greater than a first threshold. If it is, the character is assiened 590 a"*".
Any symbol may be assigned to the character. However, as the tirst threshold is the highest threshold, the symbol should visually represent the fact that the character is in the most accurate category. After assigning 590 the "*". the svstem repeats 598 for the next character. or if it was the last character. the system repeats 599 tor the next word grouping.
If it ~vas the last character in the last word grouping, the svstem proceeds to the display step ? 16.
lf the sclected character did not exceed the first threshold. the svstem determines 586 hether the character's accuracy value 250 exceeds the second tllreshold. I
t it does, then a "+" is assigned 592. Again, any symbol may be used but the symbol should represent to the user 180 that the character is in the second best category.
If the character does not pass the second threshold, the system compares 588 the character to a third threshold. If the character exceeds the third threshold. a"-" is assigned. If the character fails to exceed the third threshold, a "-" is assigned. After all of the characters have been assigned a symbol. the words and their symbols are displayed 216 to the user 180.
Word grouping accuracy values 260 generated in accordance with this method are advantageous because they visually communicate to the user 180 the quality of the characters in a given word. Thus, if a user 180 sees a word and the symbols "**+*"
displayed. the user 180 knows that the word has been recognized accurately.
This visual representation may be more effective in communicating accuracy values 260 than numeric displays. The thresholds in the above method are preferably set by the user 180 to suit the user's preferences. One of ordinary skill in the art can see that other methods ot'calculating word grouping accuracy values 260 may be used in accordance with this 1~ invention. For example, symbols may be assigned to words instead of characters, allowing the user 180 to have a visual representation of word grouping accuracy values 260. Or. the word grouping accuracy value 260 of a word may be set equal to the accuracy value 250 of the character in the word grouping which has the lowest accuracy %,alue 250.
~0 For all of the above methods, after the word grouping accuracy value 260 calculated tor a word grouping, the value and its associated word grouping are stored in a 19100-3177//709113.I
table 170 located on disk 160. The table 170 is used to retrieve the accuracy value 260 for a word grouping when the value 260 needs to be adjusted or displayed.
Figure 6a describes one embodiment of the adjusting word accuracy module 416.
In this embodiment. the word grouping accuracy value 260 of a selected word grouping is adjusted based upon comparisons of the selected word grouping to word groupings on a stored list. The stored list is preferably composed of commonly used words such as those found in a dictionary. Additionally, words or acronyms that are commonly used in the tvpes of documents beine processed by the system are preferably added to the stored list by the user 180. For example, if a company using the present invention is a software development companv. thev mav create a list including words such as "debug", "beta", and "GUI." The additional or custom words are added to the dictionary word list, or. in a preferred embodiment. the custom words are stored in a separate word list.
In accordance with the present invention, a first word grouping is selected 600.
The word grouping is compared 604 to a word grouping in the stored list. The system determines 608 whether there is a match between the selected word grouping and the word grouping on the stored list. If there is a match. a constant is added 628 to the accuracy value 260 of the selected word grouping. The system determines 632 if there is another word grouping to be compared, and if there is. a counter for the selected word groupings is increased 636. the counter for the stored list is set 640 to one, and the counter for the stored list is set 644 to one. If the system determines that there are no more word groupings to be compared, the system moves to the display module 216.
19100-3177.'/7091t3.1 If there xvas not a match between the selected word grouping and the word grouping of the stored list, the system detertnines 612 if this the last word grouping on the stored list. If there are more word groupings on the stored list, a counter is incremented 616. and the next word on the list is compared with the selected word grouping. If there are no more word groupings on the stored list, the system determines 620 if there are more stored lists. If there are more stored lists, a counter is incremented 622 and the comparison process is repeated for the new stored list. If there are no more stored lists, no value is added to the selected word grouping, and the system moves on to the next word grouping in process step 632.
In the example of the word "earring," the word grouping accuracy value 260 was 0.71. If the above module 416 were applied, "earring" would be compared to a stored list of words. I f a dictionary was one of the stored lists, a match would be found. The adjusting module 416 would increase the value of the word accuracy to .81 because of the match, using a constant of .05. The user 180 would then see the increased confidence factor, and feel more confident that the word was accurately recognized.
However, if in the original example. the word was "GUI", and the "I" was misrecognized as a"L", the word "GUL" would have been created. The OCR confidence factor for the "L" may still be high, for example, at .75, because the OCR may believe that the "L" was recognized correctly. A user 180 who sees "GUL" who is not the author of the document may be unsure w-hether GUL is a new acronym or a misrecognized word. However. a stored list comparison containing technical words and acronyms would not find a match for "GUL", and therefore. the value would not be adjusted. Since words correctly recognized will 1 9 1 00-3 1 77 ,'7091 13.1 have a .5 value added. the %alue for "GUL" %%ill appear lo%~ to the user 180. -I'he user 180 %%ill have less contidence that the word "GUL" is accuratelv recoenized.
and mav choose to exclude "GUL" from beine used to index the document. Thus, one advantaLye of this system is that it allo~ks non-experts to perform document entry. In another embodiment of the system, a constant is subtracted from the %tiord accuracy measure if no match is found. In either embodiment. the system is optimized if the authors of documents update the word lists trequently with the acronvms and special word uses of the authors.
In Figure 6b, an alternate embodiment of the adjusting -word accuracy module is illustrated. In this embodiment. language rules are applied to the word grouping to adjust the word grouping accuracy value 260 of the selected word grouping. Language rules are rules such as no capital letter in the middle of a word. For example. if an OCR
recognizes the word "example" as "examPle". the application of a language rule would catch the mistake and subtract a constant from the word grouping accuracy value 260.
Other language rules mav include "no i before e except atter c." the first letter of' the tirst word of a sentence should be capitalized, etc. The list of language rules mav be edited by the user 180 to create language rules that may be specific to the user's field. For example, if the present invention is used for a foreign language. foreign language rules may be implemented.
As illustrated in Figure 6b. a kvord grouping is selected 648. 'I'he system determines 652 if the word grouping contradicts the tirst language rule. If'it does. a value is subtracted 656 froni the accuracv value 260 of the sc:lected word grouping.
I f the %~ord grouping does not contradict the first language rule, the system determines 654 if there are more language rules. If there are, a counter is incremented 668, and the next language rule is applied. If there are no more language rules to be applied or a language rule was applied, the system determines 660 if there are more word groupings. If there are not, the process proceeds to the display module 216. The above embodiment may be used alone or in conjunction with the stored list comparisons to provide a more complete word accuracy generation system. Other types of word accuracy checks may be used in accordance with the present invention.
Display module 216 displays the recognized word groupings and their associated word grouping accuracy values 260 to the user 180. This gives the user 180 an opportunity to understand how accurately the OCR recognized the document. If the OCR
recognized the words poorly, the user 180 may be forced to delete the OCR
version, and use the paper document in its paper form. Alternatively, the user 180 may employ another OCR to obtain a better result. If the OCR repeatedly generates inaccurate word ~~roupings. the user 180 may decide to purchase another OCR system.
Keeping track of the OCR's performance is also important in an embodiment wherein the recognized word groupings are used to index a document. In this embodiment, an indexing list of the recognized word groupings is used to retrieve documents. The word groupings on the list are used as potential search terms for 10 retric%-ing the document. For example, if the word "rock" was in a word list for I)ocumcnt A. the word "rock" may be entered as a search term to find and retrieve 19100-3177//709113.1 Document A. The indexing list is formed from the table 170 of word groupings and their accuracy values 260.
To begin the process. the table 170 is displayed to the user 180. The user 180 can then determine whether or not to remove some of the word groupings from the table 170.
Words removed from the table 170 will not be on the indexing list. For example. the user 180 may want to prevent any word that does not appear to the user 180 to be an accuratelv recognized word from the list so it cannot be used to retrieve the document.
This ensures that the word list used to index the document is kept free of "pollutants,"
which are words that are misrecognized and may introduce error into the file retrieval process. For example. if the word "ear" is on the list for Document A. but the true word was "eat", entering "ear" as a search term will retrieve Document A
erroneously.
Ho-wever. if "ear" is displayed with a low word grouping accuracy value 260, the user 180 may choose to remove that word based upon its accuracy value 260 from the word list for Document A.
In the preferred embodiment, the user 180 simply specifies a level of accuracy value 260 that the user 180 would like to maintain in the user's word lists.
For example, the user 180 may specify that all words that are recognized at less than .95 should not be kept. In this embodiment, after a word grouping has its final accuracy value determined. its accuracy value 260 is compared to the threshold. If it passes, the word is kept in the indexing list. If it fails, it is not stored in the indexing list.
Then. the user 180 does not have to view the results of the word accuracy generation. and does not have to perform any subsequent nianual determinations. Again, this introduces an ease uf use to 19100-3177~ ~ 7091 13. I
the document tile indexing system. as a user 180 without expertise mav be used as the document imaging/filing clerk.
Another embodiment of the present invention displays the image of the document to the user 180 using color codes to indicate the accuracy values 260 of the words. For example. one color, such as red, is assigned to words that have accuracy values 260 over a threshold of 0.5. and blue may be assigned to words having an accuracy value 260 of less than 0.5. This allows the user 180 to view the image of the document and visually understand which words are correctly recognized and which are not. The user 180 may set the threshold to whatever level of accuracy the user 180 requires.
Figure 7a illustrates an embodiment of the present invention wherein multiple optical character recognition techniques are employed. In this embodiment. the present invention calculates word grouping accuracy values 260 of each word grouping created by each OCR . These word grouping accuracy values 260 can then be used to determine the most accurate version of the word grouping.
The system begins by receiving 700 data, and creating a bitmapped image of the paper document. An OCR program is applied 704 to the bitmapped image to recognize the characters in the image. This is repeated for each OCR present in the system. Thus, multiple versions of the original document are created.
The system then obtains 708 character accuracy values 250 for each character in the first version of the original document. Next, word groupings are created 712 f'rom the first version of the document. Initial word grouping accuracy values 260 are calculated 716 for each created word grouping. The initial word grouping accuracy values 260 are 1 9 1 00-3 1 77//709 1 1 3 .1 calculated in accordance with the process described above in relation to Figures 5a-d.
The %%ord groupings and their accuracv %alues 260 are stored in a table 170.
The process is repeated 720. 722 for each OCR version. Each OCR version has its okti-n table 170 composed of word groupings and accuracy values 260.
The svstem then compares the different versions of each ~.vord grouping together to create tables 850 of versions of the word groupings ordered bv accuracv.
After an initial ordering has been created, the accuracy values 260 for each version of the v,-ord grouping are adjusted 728 in accordance %vith predefined accuracv factors.
Then. the table 850 is reordered based on the new accuracy values 260.
In one embodiment, composite word groupings are created 732. Composite word groupings are word groupings created from the most accurate versions of each character in the word grouping. Thus. tor example, if the word "help" %vas recognized by the present invention, each OCR would generate character accuracy values 250 for each of the four characters. OCR A mav generate "h": .87 "e": .92 "1": .95 "p": .77 and OCR B
mav eenerate "h": .91 "e": .90 "I": .90 "p": .82. Therefore. if OCR A and B-,kere the onlv OCRs present in the system, a composite ord grouping is created using the "h"
from OCR B. the "e" from OCR A. the "I" from OCR A. and the "p" from OCR B.
Finally, the most accurate word groupings 736 are displayed or stored into an indexing list. If composite %tiord groupings are used. a word grouping accuracv value 260 is calcu(ated for the composite word grouping and is compared with the other ,kord groupings to determine the most accurate word groupings.
, ~
_~
19 1 00-3 1 77." ',091 13 .1 Figure 8 illustrates ordering the versions of the word groupings in greater detail.
A word grouping is selected 804 from the table 170 of word groupings created by the first OCR. The coordinates of the selected word grouping on the bitmapped image are determined 808. This process is performed in order to ensure that versions of the same ~ word grouping are compared against each other. The version of the word grouping is then stored 812 in a first word grouping version table 850. A counter is incremented 816, and a next table 170 of word groupings created by a next OCR is selected 820.
A word grouping is selected 824 that corresponds to the coordinates determined previously. The accuracv value 260 of the selected version is compared 828 to the accuracy value 260 of the version of the word grouping already stored in the word grouping version table 850.
The system then determines 832 whether the selected version has a greater accuracy value 260 than the version in the table 850. If it has, the selected version is stored 844 in the table 850 in a position immediately prior to the first version. If the selected version has a lesser accuracy value 260, the system determines 836 if there are more versions of the word grouping stored in the word grouping version table 850. If there are not, the selected version is stored 848 in a position immediately after the first x=ersion. If there are multiple versions of the word grouping already stored in the word grouping version table 850, the selected version is compared against all of the versions in the table 850 until a version is found that has a greater accuracy value 260.
At that point, '0 the ~. ersion is stored in the table 850 in a position after the version having a greater accuracy value 260. For example, as illustrated in Figure 8b, if the word "example" was in the original paper document, and there were OCRs A. B. C, and D present in the 19100-3177//7091 13 .1 system, there would be four versions of the word "example" created. The system creates a table 850 of versions of the word "example." and orders the versions by accuracv.
Thus. as shown. the calculated word grouping accuracy values 260 for the versions of example are: OCR A"example": .91. OCR B "example": .87. OCR C "example": .72 and OCR D "example": .95. The table 850 therefore orders the versions beginning with OCR
D's "example". followed bv A's, B's. and C's. If a fifth version of "example"
is created by a fifth OCR. the fifth version would be compared to the four versions already in the table 850. The accuracy value of the fifth version is compared to the accuracy values of the other four. and is placed into the table accordingly. If the fifth version has an accuracy value 260 of.89. its version is placed in a position under OCR A's version and above OCR D's version.
The system then determines 852 if there are more OCR versions of the selected -ord Llrouping. [f there are. a next version of the word grouping is selected, and the process is repeated. If there are no more versions of the word grouping, the system determines if there are more word groupings in the first OCR word grouping table 170. If there are. the next word grouping is selected, and the process is repeated.
Figure 9 illustrates adjusting accuracy values 260 after the versions of each word grouping have been ordered. First, a factor or constant is added 900 to a version of a word grouping based upon the strengths of the OCR which created the version of the word grouping. For example. if the word "help" has been created by OCR A. and OCR
A recoenizes Courier font particularly well, the word grouping accuracy value 260 for the version of help created by OCR A is adjusted upwards by a constant. This is repeated for '5 19100-3177,'/ 7091 13.1 each version of the word grouping in the word grouping table 170. Then the versions of the word grouping on the word grouping table 170 are reordered 904 based upon the changes in accuracy values 260.
A factor or constant is subtracted 908 from the versions based upon OCR weaknesses in the manner described above. The table 170 is reordered 912 based upon any change in accuracy values 260. Then, the word grouping accuracy values 260 are adjusted and reordered based upon stored list comparisons 916 and language rule applications 920, as described in Figures 6a and 6b. The adjusting and reordering produces a final ordering of word grouping versions from most accurate to least accurate. At this point, all word groupings having an accuracy value over a threshold may be kept, or only the most .15 accurate word grouping version may be kept.
Figure 10 illustrates the display word grouping aspect of the present invention, which may be used in a single or multiple OCR embodiments of the present invention. However, it is described here with respect to the multiple OCR
embodiment of the present invention. A first word grouping version table 850 is selected 1000. The version in the first position in the table 850 is retrieved as it is the most accurate version based upon the ordering performed earlier.
In one embodiment, the composite version of the word grouping is compared to the retrieved version to determine which is the most accurate version. This embodiment may be made into an optional step dependent upon the accuracy of the most accurate version of the word grouping. If the most accurate version of the word grouping has an accuracy value 260 greater than a threshold, no composite word grouping is created. This saves on processing time, as composite word grouping creating may be a time-consuming process for low-speed processors. The threshold may be specified by the system, or is preferably a user-defined threshold level.
Finally, the most accurate version of the word grouping is displayed. Alternatively, the most accurate version is used to create a list used for indexing the document, as shown in Figure 11. In this embodiment, if the system determines 1100 that a version of the word grouping has an accuracy value which exceeds a threshold, the version is used to update 1104 an indexing list. Again, the threshold is a level of accuracy the user 180 desires for the user's system. By keeping all of the versions which exceed the threshold, the user 180 can eliminate, to the greatest extent possible, OCR mistakes which are not caught by the above-described processes. For example, if the table 850 contains three versions of the word "ear"
which exceed the threshold, and the most accurate version is "eat", and the other two are "ear", a system which keeps the version with the highest accuracy value will incorrectly keep "eat" as the version of the word. A system which keeps all three will also keep the versions which have the correct representation of the word "ear." Thus, in those systems, the document is able to be retrieved by using "ear" as a search term. However, by keeping multiple versions, more disk space is used in keeping the longer indexing lists.
If none of the word groupings exceed the threshold, in either the single or multiple OCR embodiments of the present invention, the user 180 is preferably displayed 1108 several options regarding how to proceed. After receiving 1112 an input, the system executes the option selected. A first option 1116 is to perform the OCR again, in an attempt to increase the character accuracy values. If multiple OCRs are present in the svstem, the user 180 may select one of the other OCRs. A second option 1120 is to re-scan the document. This may provide a cleaner image for the OCR to recognize.
A third option 1124 is to allow the user 180 to lower the threshold. If the word groupings consistentlv fail to meet the threshold, the user 180 may have set the threshold too high for the user's svstem.
The above description is included to illustrate the operation of the preferred embodiments and is not meant to limit the scope of the invention. The scope of the invention is to be limited only by the following claims. From the above discussion, many variations will be apparent to one skilled in the art that would yet be encompassed by the spirit and scope of the present invention.
What is claimed is:
19100-3177//7091 13.1
Claims (19)
1. A method for calculating word accuracy values, comprising:
recognizing characters within image data by performing a character recognition technique;
obtaining character accuracy values from the recognized characters;
creating words from the recognized characters;
selecting one of the created words;
calculating a word accuracy value of the selected word on the basis of the character accuracy values obtained on said obtaining step;
discriminating an accuracy level of the selected word from a plurality of accuracy level by comparing the calculated word accuracy value to a predetermined threshold value, wherein each accuracy level being associated with a visual identifier;
assigning the visual identifier associated with the discriminated accuracy level to the selected word; and repeating the selecting, calculating, discriminating, and assigning steps for each created word.
recognizing characters within image data by performing a character recognition technique;
obtaining character accuracy values from the recognized characters;
creating words from the recognized characters;
selecting one of the created words;
calculating a word accuracy value of the selected word on the basis of the character accuracy values obtained on said obtaining step;
discriminating an accuracy level of the selected word from a plurality of accuracy level by comparing the calculated word accuracy value to a predetermined threshold value, wherein each accuracy level being associated with a visual identifier;
assigning the visual identifier associated with the discriminated accuracy level to the selected word; and repeating the selecting, calculating, discriminating, and assigning steps for each created word.
2. The method according to claim 1, wherein said image data includes digital representations of text or graphics symbols.
3. The method according to claim 1, wherein said calculating step comprises the sub-steps of:
selecting a character within the selected word;
determining whether the character accuracy value of the selected character exceeds a threshold value;
responsive to the character accuracy value exceeding the threshold value, assigning a"one" to the character;
responsive to determining that the character accuracy value does not exceed the threshold value, assigning a "zero" to the character;
repeating the selecting, determining, assigning a "one", and assigning a "zero" sub-steps for each character in the selected word; and determining a word accuracy value by the logical combination of characters assigned a "one" and the total number of characters in the selected word.
selecting a character within the selected word;
determining whether the character accuracy value of the selected character exceeds a threshold value;
responsive to the character accuracy value exceeding the threshold value, assigning a"one" to the character;
responsive to determining that the character accuracy value does not exceed the threshold value, assigning a "zero" to the character;
repeating the selecting, determining, assigning a "one", and assigning a "zero" sub-steps for each character in the selected word; and determining a word accuracy value by the logical combination of characters assigned a "one" and the total number of characters in the selected word.
4. The method according to claim 3, wherein said word accuracy value determining sub-step determines the word accuracy value by dividing the number of characters assigned a "one" by the total number of characters in the selected word.
5. The method according to claim 1, further comprising the steps of:
decreasing the obtained character accuracy values responsive to the recognized character being a character known to be less accurately recognized by the character recognition technique; and increasing the obtained character accuracy values responsive to the recognized character being a character known to be more accurately recognized by the character recognition technique.
decreasing the obtained character accuracy values responsive to the recognized character being a character known to be less accurately recognized by the character recognition technique; and increasing the obtained character accuracy values responsive to the recognized character being a character known to be more accurately recognized by the character recognition technique.
6. The method according to claim 1, further comprising the step of decreasing the calculated word accuracy value responsive to the selected word contradicting one of predetermined language rules.
7. The method according to claim 1, further comprising the steps of:
increasing the calculated word accuracy value responsive to the selected word matching one of the words in a list of words, the list being stored in advance; and decreasing the calculated word accuracy value responsive to the selected word not matching one of the words in the list.
increasing the calculated word accuracy value responsive to the selected word matching one of the words in a list of words, the list being stored in advance; and decreasing the calculated word accuracy value responsive to the selected word not matching one of the words in the list.
8. The method according to claim 7, wherein the stored lists of words include lists of technical words, words found in a dictionary, foreign words, and trade words.
9. The method according to claim 1, further comprising the step of displaying the created words and the assigned visual identifiers for each word.
10. The method according to claim 1, wherein said calculating step comprises the sub-steps of:
responsive to determining that all of the character accuracy values are at least equal to a threshold accuracy value, calculating the word accuracy value of the selected word by calculating the average of the character accuracy values;
and responsive to determining that one or more of the character accuracy values are less than the threshold accuracy value, calculating the word accuracy value by dividing the number of characters that do not have an accuracy value that exceeds the threshold by one hundred, and by subtracting the dividing result from the threshold accuracy value.
responsive to determining that all of the character accuracy values are at least equal to a threshold accuracy value, calculating the word accuracy value of the selected word by calculating the average of the character accuracy values;
and responsive to determining that one or more of the character accuracy values are less than the threshold accuracy value, calculating the word accuracy value by dividing the number of characters that do not have an accuracy value that exceeds the threshold by one hundred, and by subtracting the dividing result from the threshold accuracy value.
11. The method according to claim 1, wherein said calculating step comprises the sub-steps of:
determining the minimum character accuracy value from the obtained character accuracy values of characters contained in the selected words; and setting the word accuracy value of the selected word equal to the determined minimum character accuracy value.
determining the minimum character accuracy value from the obtained character accuracy values of characters contained in the selected words; and setting the word accuracy value of the selected word equal to the determined minimum character accuracy value.
12. The method according to claim 1, further comprising the step of:
adding to an indexing list word whose accuracy values exceed a threshold accuracy level, the indexing list being used for searching and retrieving documents.
adding to an indexing list word whose accuracy values exceed a threshold accuracy level, the indexing list being used for searching and retrieving documents.
13. The method according to claim 12, further comprising the step of:
responsive to none of the word having an accuracy value which exceeds the threshold accuracy level, displaying an option to set the threshold accuracy level to a different value.
responsive to none of the word having an accuracy value which exceeds the threshold accuracy level, displaying an option to set the threshold accuracy level to a different value.
14. The method according to claim 1, wherein said recognizing step recognizes characters by performing a plurality of character recognition techniques, and wherein said obtaining step obtains character accuracy values for each character recognition technique, and wherein said creating step creates words for each character recognition technique, and wherein said calculating step calculates a word accuracy value of the created words for each character recognition technique, and wherein said method further comprises the step of determining the most accurate words based on the calculated word accuracy values, and wherein said discriminating and assigning steps are performed for the most accurate words.
15. The method according to claim 1, wherein said visual identifiers are visual quality symbols.
16. The method according to claim 1, wherein said visual identifiers are color codes.
17. The method according to claim 1, wherein said discriminating step discriminates accuracy levels of each character of the selected word, and wherein said assigning step assigns the visual identifier associated with the discriminated accuracy levels to each character of the selected word.
18. A computer-readable storage medium having stored thereon statements or instructions, for execution by a computer, for carrying out the steps of any one of claims 1 to 17.
19. An apparatus for calculating word accuracy values, comprising:
recognizing means for recognizing characters within image data by performing a character recognition technique;
obtaining means for obtaining character accuracy values from the recognized characters;
creating means for creating words from the recognized characters;
selecting means for selecting one of the created words; calculating means for calculating a word accuracy value of the selected word on the basis of the character accuracy values obtained by said obtaining means;
discriminating means for discriminating an accuracy level of the selected word from a plurality of accuracy level by comparing the calculated word accuracy value to a predetermined threshold value, wherein each accuracy level being associated with a visual identifier;
assigning means for assigning the visual identifier associated with the discriminated accuracy level to the selected word; and repeating means for repeating the processes by the selecting, calculating, discriminating, and assigning means for each created words.
recognizing means for recognizing characters within image data by performing a character recognition technique;
obtaining means for obtaining character accuracy values from the recognized characters;
creating means for creating words from the recognized characters;
selecting means for selecting one of the created words; calculating means for calculating a word accuracy value of the selected word on the basis of the character accuracy values obtained by said obtaining means;
discriminating means for discriminating an accuracy level of the selected word from a plurality of accuracy level by comparing the calculated word accuracy value to a predetermined threshold value, wherein each accuracy level being associated with a visual identifier;
assigning means for assigning the visual identifier associated with the discriminated accuracy level to the selected word; and repeating means for repeating the processes by the selecting, calculating, discriminating, and assigning means for each created words.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US09/041,854 US6269188B1 (en) | 1998-03-12 | 1998-03-12 | Word grouping accuracy value generation |
| US09/041,854 | 1998-03-12 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CA2265060A1 CA2265060A1 (en) | 1999-09-12 |
| CA2265060C true CA2265060C (en) | 2007-05-22 |
Family
ID=21918699
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002265060A Expired - Fee Related CA2265060C (en) | 1998-03-12 | 1999-03-08 | Word grouping accuracy value generation |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US6269188B1 (en) |
| EP (1) | EP0942389B1 (en) |
| JP (1) | JPH11316800A (en) |
| CA (1) | CA2265060C (en) |
| DE (1) | DE69940331D1 (en) |
Families Citing this family (73)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8352400B2 (en) | 1991-12-23 | 2013-01-08 | Hoffberg Steven M | Adaptive pattern recognition based controller apparatus and method and human-factored interface therefore |
| US7966078B2 (en) | 1999-02-01 | 2011-06-21 | Steven Hoffberg | Network media appliance system and method |
| JP3913985B2 (en) * | 1999-04-14 | 2007-05-09 | 富士通株式会社 | Character string extraction apparatus and method based on basic components in document image |
| CN1411586A (en) * | 2000-03-06 | 2003-04-16 | 埃阿凯福斯公司 | System and method for creating searchable word index of scanned document including multiple interpretations of word at given document location |
| US7155061B2 (en) * | 2000-08-22 | 2006-12-26 | Microsoft Corporation | Method and system for searching for words and phrases in active and stored ink word documents |
| JP4136316B2 (en) * | 2001-01-24 | 2008-08-20 | 富士通株式会社 | Character string recognition device |
| US7146321B2 (en) * | 2001-10-31 | 2006-12-05 | Dictaphone Corporation | Distributed speech recognition system |
| US7133829B2 (en) * | 2001-10-31 | 2006-11-07 | Dictaphone Corporation | Dynamic insertion of a speech recognition engine within a distributed speech recognition system |
| US6785654B2 (en) | 2001-11-30 | 2004-08-31 | Dictaphone Corporation | Distributed speech recognition system with speech recognition engines offering multiple functionalities |
| US6766294B2 (en) | 2001-11-30 | 2004-07-20 | Dictaphone Corporation | Performance gauge for a distributed speech recognition system |
| US20030128856A1 (en) * | 2002-01-08 | 2003-07-10 | Boor Steven E. | Digitally programmable gain amplifier |
| US7236931B2 (en) | 2002-05-01 | 2007-06-26 | Usb Ag, Stamford Branch | Systems and methods for automatic acoustic speaker adaptation in computer-assisted transcription systems |
| US7292975B2 (en) * | 2002-05-01 | 2007-11-06 | Nuance Communications, Inc. | Systems and methods for evaluating speaker suitability for automatic speech recognition aided transcription |
| MXPA04011507A (en) * | 2002-05-20 | 2005-09-30 | Tata Infotech Ltd | Document structure identifier. |
| US7171061B2 (en) * | 2002-07-12 | 2007-01-30 | Xerox Corporation | Systems and methods for triage of passages of text output from an OCR system |
| US7045377B2 (en) * | 2003-06-26 | 2006-05-16 | Rj Mears, Llc | Method for making a semiconductor device including a superlattice and adjacent semiconductor layer with doped regions defining a semiconductor junction |
| WO2005009205A2 (en) * | 2003-07-09 | 2005-02-03 | Gensym Corporation | System and method for self management of health using natural language interface |
| US7707039B2 (en) | 2004-02-15 | 2010-04-27 | Exbiblio B.V. | Automatic modification of web pages |
| US8442331B2 (en) | 2004-02-15 | 2013-05-14 | Google Inc. | Capturing text from rendered documents using supplemental information |
| JP4297798B2 (en) * | 2004-01-29 | 2009-07-15 | 富士通株式会社 | Mobile information management program |
| US20060104515A1 (en) * | 2004-07-19 | 2006-05-18 | King Martin T | Automatic modification of WEB pages |
| US10635723B2 (en) | 2004-02-15 | 2020-04-28 | Google Llc | Search engines and systems with handheld document data capture devices |
| US7812860B2 (en) | 2004-04-01 | 2010-10-12 | Exbiblio B.V. | Handheld device for capturing text from both a document printed on paper and a document displayed on a dynamic display device |
| US20060136629A1 (en) * | 2004-08-18 | 2006-06-22 | King Martin T | Scanner having connected and unconnected operational behaviors |
| US7552630B2 (en) * | 2004-02-27 | 2009-06-30 | Akron Special Machinery, Inc. | Load wheel drive |
| US9143638B2 (en) | 2004-04-01 | 2015-09-22 | Google Inc. | Data capture from rendered documents using handheld device |
| US7990556B2 (en) | 2004-12-03 | 2011-08-02 | Google Inc. | Association of a portable scanner with input/output and storage devices |
| US9116890B2 (en) | 2004-04-01 | 2015-08-25 | Google Inc. | Triggering actions in response to optically or acoustically capturing keywords from a rendered document |
| US8146156B2 (en) | 2004-04-01 | 2012-03-27 | Google Inc. | Archive of text captures from rendered documents |
| US20060098900A1 (en) | 2004-09-27 | 2006-05-11 | King Martin T | Secure data gathering from rendered documents |
| US8081849B2 (en) | 2004-12-03 | 2011-12-20 | Google Inc. | Portable scanning and memory device |
| US20060081714A1 (en) | 2004-08-23 | 2006-04-20 | King Martin T | Portable scanning device |
| US7894670B2 (en) | 2004-04-01 | 2011-02-22 | Exbiblio B.V. | Triggering actions in response to optically or acoustically capturing keywords from a rendered document |
| US9008447B2 (en) | 2004-04-01 | 2015-04-14 | Google Inc. | Method and system for character recognition |
| US8713418B2 (en) | 2004-04-12 | 2014-04-29 | Google Inc. | Adding value to a rendered document |
| US8489624B2 (en) | 2004-05-17 | 2013-07-16 | Google, Inc. | Processing techniques for text capture from a rendered document |
| US8874504B2 (en) | 2004-12-03 | 2014-10-28 | Google Inc. | Processing techniques for visual capture data from a rendered document |
| US8620083B2 (en) | 2004-12-03 | 2013-12-31 | Google Inc. | Method and system for character recognition |
| US8346620B2 (en) | 2004-07-19 | 2013-01-01 | Google Inc. | Automatic modification of web pages |
| US8032372B1 (en) | 2005-09-13 | 2011-10-04 | Escription, Inc. | Dictation selection |
| US7734092B2 (en) * | 2006-03-07 | 2010-06-08 | Ancestry.Com Operations Inc. | Multiple image input for optical character recognition processing systems and methods |
| US7966557B2 (en) | 2006-03-29 | 2011-06-21 | Amazon Technologies, Inc. | Generating image-based reflowable files for rendering on various sized displays |
| EP2067119A2 (en) | 2006-09-08 | 2009-06-10 | Exbiblio B.V. | Optical scanners, such as hand-held optical scanners |
| US7810026B1 (en) * | 2006-09-29 | 2010-10-05 | Amazon Technologies, Inc. | Optimizing typographical content for transmission and display |
| US8595615B2 (en) * | 2007-02-07 | 2013-11-26 | International Business Machines Corporation | System and method for automatic stylesheet inference |
| US8782516B1 (en) | 2007-12-21 | 2014-07-15 | Amazon Technologies, Inc. | Content style detection |
| JP2009193356A (en) * | 2008-02-14 | 2009-08-27 | Canon Inc | Image processing apparatus, image processing method, program, and storage medium |
| US20090300126A1 (en) * | 2008-05-30 | 2009-12-03 | International Business Machines Corporation | Message Handling |
| US8572480B1 (en) | 2008-05-30 | 2013-10-29 | Amazon Technologies, Inc. | Editing the sequential flow of a page |
| US9229911B1 (en) * | 2008-09-30 | 2016-01-05 | Amazon Technologies, Inc. | Detecting continuation of flow of a page |
| DE202010018601U1 (en) | 2009-02-18 | 2018-04-30 | Google LLC (n.d.Ges.d. Staates Delaware) | Automatically collecting information, such as gathering information using a document recognizing device |
| US8447066B2 (en) | 2009-03-12 | 2013-05-21 | Google Inc. | Performing actions based on capturing information from rendered documents, such as documents under copyright |
| WO2010105245A2 (en) | 2009-03-12 | 2010-09-16 | Exbiblio B.V. | Automatically providing content associated with captured information, such as information captured in real-time |
| US20100274615A1 (en) * | 2009-04-22 | 2010-10-28 | Eran Belinsky | Extendable Collaborative Correction Framework |
| US9135277B2 (en) | 2009-08-07 | 2015-09-15 | Google Inc. | Architecture for responding to a visual query |
| US8670597B2 (en) | 2009-08-07 | 2014-03-11 | Google Inc. | Facial recognition with social network aiding |
| US9087059B2 (en) | 2009-08-07 | 2015-07-21 | Google Inc. | User interface for presenting search results for multiple regions of a visual query |
| US20110099193A1 (en) * | 2009-10-26 | 2011-04-28 | Ancestry.Com Operations Inc. | Automatic pedigree corrections |
| US8600152B2 (en) * | 2009-10-26 | 2013-12-03 | Ancestry.Com Operations Inc. | Devices, systems and methods for transcription suggestions and completions |
| US8805079B2 (en) | 2009-12-02 | 2014-08-12 | Google Inc. | Identifying matching canonical documents in response to a visual query and in accordance with geographic information |
| US8811742B2 (en) | 2009-12-02 | 2014-08-19 | Google Inc. | Identifying matching canonical documents consistent with visual query structural information |
| US9176986B2 (en) | 2009-12-02 | 2015-11-03 | Google Inc. | Generating a combination of a visual query and matching canonical document |
| US9405772B2 (en) | 2009-12-02 | 2016-08-02 | Google Inc. | Actionable search results for street view visual queries |
| US9183224B2 (en) | 2009-12-02 | 2015-11-10 | Google Inc. | Identifying matching canonical documents in response to a visual query |
| US8977639B2 (en) | 2009-12-02 | 2015-03-10 | Google Inc. | Actionable search results for visual queries |
| US9852156B2 (en) | 2009-12-03 | 2017-12-26 | Google Inc. | Hybrid use of location sensor data and visual query to return local listings for visual query |
| US9081799B2 (en) | 2009-12-04 | 2015-07-14 | Google Inc. | Using gestalt information to identify locations in printed information |
| US9323784B2 (en) | 2009-12-09 | 2016-04-26 | Google Inc. | Image search using text-based elements within the contents of images |
| US8499236B1 (en) | 2010-01-21 | 2013-07-30 | Amazon Technologies, Inc. | Systems and methods for presenting reflowable content on a display |
| AU2011336445B2 (en) * | 2010-12-01 | 2017-04-13 | Google Llc | Identifying matching canonical documents in response to a visual query |
| US9383913B2 (en) * | 2012-05-30 | 2016-07-05 | Sap Se | Touch screen device data filtering |
| US8935246B2 (en) | 2012-08-08 | 2015-01-13 | Google Inc. | Identifying textual terms in response to a visual query |
| RU2634194C1 (en) * | 2016-09-16 | 2017-10-24 | Общество с ограниченной ответственностью "Аби Девелопмент" | Verification of optical character recognition results |
Family Cites Families (37)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3969698A (en) * | 1974-10-08 | 1976-07-13 | International Business Machines Corporation | Cluster storage apparatus for post processing error correction of a character recognition machine |
| US4941125A (en) | 1984-08-01 | 1990-07-10 | Smithsonian Institution | Information storage and retrieval system |
| US5265242A (en) | 1985-08-23 | 1993-11-23 | Hiromichi Fujisawa | Document retrieval system for displaying document image data with inputted bibliographic items and character string selected from multiple character candidates |
| JP2695844B2 (en) | 1988-06-16 | 1998-01-14 | 株式会社東芝 | Document shaping device |
| DE68913669T2 (en) * | 1988-11-23 | 1994-07-21 | Digital Equipment Corp | Pronunciation of names by a synthesizer. |
| EP0380239A3 (en) | 1989-01-18 | 1992-04-15 | Lotus Development Corporation | Search and retrieval system |
| JP2816241B2 (en) | 1990-06-20 | 1998-10-27 | 株式会社日立製作所 | Image information retrieval device |
| US5757983A (en) | 1990-08-09 | 1998-05-26 | Hitachi, Ltd. | Document retrieval method and system |
| JP3303926B2 (en) | 1991-09-27 | 2002-07-22 | 富士ゼロックス株式会社 | Structured document classification apparatus and method |
| US5926565A (en) | 1991-10-28 | 1999-07-20 | Froessl; Horst | Computer method for processing records with images and multiple fonts |
| US5875263A (en) | 1991-10-28 | 1999-02-23 | Froessl; Horst | Non-edit multiple image font processing of records |
| US5375235A (en) | 1991-11-05 | 1994-12-20 | Northern Telecom Limited | Method of indexing keywords for searching in a database recorded on an information recording medium |
| JP2579397B2 (en) | 1991-12-18 | 1997-02-05 | インターナショナル・ビジネス・マシーンズ・コーポレイション | Method and apparatus for creating layout model of document image |
| US5359667A (en) | 1992-08-24 | 1994-10-25 | Unisys Corporation | Method for identifying and tracking document characteristics in a document image processing system |
| US6002798A (en) | 1993-01-19 | 1999-12-14 | Canon Kabushiki Kaisha | Method and apparatus for creating, indexing and viewing abstracted documents |
| US5848184A (en) | 1993-03-15 | 1998-12-08 | Unisys Corporation | Document page analyzer and method |
| JP3302147B2 (en) | 1993-05-12 | 2002-07-15 | 株式会社リコー | Document image processing method |
| CN1045679C (en) * | 1993-12-01 | 1999-10-13 | 摩托罗拉公司 | Combined dictionary based and likely character string method of handwriting recognition |
| EP0667594A3 (en) | 1994-02-14 | 1995-08-23 | International Business Machines Corporation | Image quality analysis method and apparatus |
| CA2144793C (en) | 1994-04-07 | 1999-01-12 | Lawrence Patrick O'gorman | Method of thresholding document images |
| JPH087033A (en) * | 1994-06-16 | 1996-01-12 | Canon Inc | Method and device for processing information |
| US5802205A (en) | 1994-09-09 | 1998-09-01 | Motorola, Inc. | Method and system for lexical processing |
| US5675665A (en) * | 1994-09-30 | 1997-10-07 | Apple Computer, Inc. | System and method for word recognition using size and placement models |
| JP3669016B2 (en) | 1994-09-30 | 2005-07-06 | 株式会社日立製作所 | Document information classification device |
| US5805747A (en) * | 1994-10-04 | 1998-09-08 | Science Applications International Corporation | Apparatus and method for OCR character and confidence determination using multiple OCR devices |
| JP3647518B2 (en) | 1994-10-06 | 2005-05-11 | ゼロックス コーポレイション | Device that highlights document images using coded word tokens |
| US5642288A (en) | 1994-11-10 | 1997-06-24 | Documagix, Incorporated | Intelligent document recognition and handling |
| JP3375766B2 (en) * | 1994-12-27 | 2003-02-10 | 松下電器産業株式会社 | Character recognition device |
| US5617488A (en) * | 1995-02-01 | 1997-04-01 | The Research Foundation Of State University Of New York | Relaxation word recognizer |
| US5764799A (en) * | 1995-06-26 | 1998-06-09 | Research Foundation Of State Of State Of New York | OCR method and apparatus using image equivalents |
| US5781879A (en) * | 1996-01-26 | 1998-07-14 | Qpl Llc | Semantic analysis and modification methodology |
| US5850480A (en) | 1996-05-30 | 1998-12-15 | Scan-Optics, Inc. | OCR error correction methods and apparatus utilizing contextual comparison |
| JP2973944B2 (en) | 1996-06-26 | 1999-11-08 | 富士ゼロックス株式会社 | Document processing apparatus and document processing method |
| US5933531A (en) * | 1996-08-23 | 1999-08-03 | International Business Machines Corporation | Verification and correction method and system for optical character recognition |
| US5878385A (en) | 1996-09-16 | 1999-03-02 | Ergo Linguistic Technologies | Method and apparatus for universal parsing of language |
| US6006226A (en) | 1997-09-24 | 1999-12-21 | Ricoh Company Limited | Method and system for document image feature extraction |
| US5999664A (en) | 1997-11-14 | 1999-12-07 | Xerox Corporation | System for searching a corpus of document images by user specified document layout components |
-
1998
- 1998-03-12 US US09/041,854 patent/US6269188B1/en not_active Expired - Lifetime
-
1999
- 1999-02-12 EP EP99301021A patent/EP0942389B1/en not_active Expired - Lifetime
- 1999-02-12 DE DE69940331T patent/DE69940331D1/en not_active Expired - Fee Related
- 1999-03-08 CA CA002265060A patent/CA2265060C/en not_active Expired - Fee Related
- 1999-03-12 JP JP11067137A patent/JPH11316800A/en active Pending
Also Published As
| Publication number | Publication date |
|---|---|
| EP0942389A3 (en) | 2000-09-20 |
| EP0942389B1 (en) | 2009-01-21 |
| CA2265060A1 (en) | 1999-09-12 |
| JPH11316800A (en) | 1999-11-16 |
| US6269188B1 (en) | 2001-07-31 |
| EP0942389A2 (en) | 1999-09-15 |
| DE69940331D1 (en) | 2009-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2265060C (en) | Word grouping accuracy value generation | |
| JP2973944B2 (en) | Document processing apparatus and document processing method | |
| JP3427692B2 (en) | Character recognition method and character recognition device | |
| Kanai et al. | Automated evaluation of OCR zoning | |
| US8731300B2 (en) | Handwritten word spotter system using synthesized typed queries | |
| US7047238B2 (en) | Document retrieval method and document retrieval system | |
| US6272242B1 (en) | Character recognition method and apparatus which groups similar character patterns | |
| US6047251A (en) | Automatic language identification system for multilingual optical character recognition | |
| EP1952285B1 (en) | System and method for searching and matching data having ideogrammatic content | |
| US5883986A (en) | Method and system for automatic transcription correction | |
| Sibun et al. | Language determination: Natural language processing from scanned document images | |
| Xu et al. | Prototype extraction and adaptive OCR | |
| US20030099399A1 (en) | Automated search on cursive records not having an ASCII index | |
| US6385339B1 (en) | Collaborative learning system and pattern recognition method | |
| Zramdini et al. | Optical font recognition from projection profiles | |
| JPH11203415A (en) | Device and method for preparing similar pattern category discrimination dictionary | |
| US20110229036A1 (en) | Method and apparatus for text and error profiling of historical documents | |
| JP4678712B2 (en) | Language identification device, program, and recording medium | |
| Hochberg et al. | Page segmentation using script identification vectors: A first look | |
| JP6533395B2 (en) | Character search method and system | |
| JPH07271921A (en) | Character recognizing device and method thereof | |
| KR100292352B1 (en) | Data editing method for recognizer by using morpheme analysis | |
| JPH096920A (en) | Method and device for handwritten character recognition | |
| JP2974145B2 (en) | Correcting character recognition results | |
| JP2827066B2 (en) | Post-processing method for character recognition of documents with mixed digit strings |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| MKLA | Lapsed |
Effective date: 20170308 |