WO2001047278A2 - Methods and systems for generating multimedia signature - Google Patents
Methods and systems for generating multimedia signature Download PDFInfo
- Publication number
- WO2001047278A2 WO2001047278A2 PCT/US2000/034803 US0034803W WO0147278A2 WO 2001047278 A2 WO2001047278 A2 WO 2001047278A2 US 0034803 W US0034803 W US 0034803W WO 0147278 A2 WO0147278 A2 WO 0147278A2
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- multimedia content
- invariant features
- codewords
- signature
- multimedia
- Prior art date
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/46—Embedding additional information in the video signal during the compression process
- H04N19/467—Embedding additional information in the video signal during the compression process characterised by the embedded information being invisible, e.g. watermarking
Definitions
- the present invention relates to techniques for generating an electronic signature for original multimedia content, and for verifying at least a portion of received multimedia content which incorporate electronic signatures.
- the sender provides a digital signature based on the content, or embeds a watermark into the content, so that the recipient can verify the originality of the content which is later received.
- the '249 patent proposes a method for invisibly watermarking high-quality color and gray-scale images for authentication purposes, and includes both a watermark stamping process which embeds a watermark in a source image, and a watermark extraction process which extracts a watermark from the stamped source image.
- the extracted watermark can be used to determine, based on complete verification, whether the source image has been altered. This method can be used for applications such as medical imaging where a change of even one pixel in the received image cannot be allowed.
- U.S. Patent 5,499,294 (hereinafter "the '294 patent”).
- the '294 patent proposes embedding an encrypted digital signature into a captured image.
- the scheme is called “trustworthy digital camera,” and is based on Public Key Encryption.
- the trustworthy digital camera scheme provides a way to protect a source image by verifying the signature to determine whether the image has been forged. If there is even a 1-bit change in the image, the verification fails. While this method allows a determination of whether an image has been altered, it cannot locate any alterations in the image.
- U.S. Patent 5,465,299 discloses a technique for generating a digital signature for tracing document changes.
- a terminal which receives an electronic document with at least one digital signature
- a new digital signature is produced by using a secret key of the user to encipher new signature data.
- the signature data includes a hash total of the document of a new version, personal information of the user, and version management information necessary to restore an electronic document of a former version from the electronic document of the new version.
- the new digital signature and the new version of the electronic document are transmitted together with the received digital signature to another person on a document circulating route.
- the invention is related only to digital contents, namely, binary based data. If the document is printed out, verification will fail.
- Authentication signatures are based on the invariance of the relationship between Discrete Cosign Transform coefficients of the same position in separate blocks of an image. This relationship is preserved when these coefficients are quantized in a JPEG compression process. Compression-allowed verification is a practical requirement in storage and transmission. However, it cannot resolve the problems of secure multimedia distribution such as format conversion, scaling and slight distortions.
- an authorized user can login into a multimedia service center, read and print some secure documents related to its business, and later present the printed documents to support its business. Another authorized viewer can also login into the server to verify the legitimacy of the printout and check whether the print-out has been altered.
- certain distortions may be present in the received content. Some examples of such distortions include scaling, color or intensity changes, slight rotation, or addition of noise. All these distortions may be considered unacceptable in existing verification methods.
- M-commerce Multimedia commerce
- An object of the present invention is to provide a technique for robustly authenticating multimedia content. Another object of the present invention is to provide a verification technique which permits high verification accuracy of multimedia content.
- Yet another object of the present invention is to provide a highly flexible verification technique.
- the present invention provides techniques for robustly and accurately authenticating multimedia content.
- robustness is advantageously achieved through a feature-based registration process, an optional error control coding process, and source-optimized vector quantization.
- Accuracy is achieved by using a typical digital signature scheme and a hierarchical verification procedure. Flexibility is advantageously achieved through scalability of vector quantization.
- invariant features of original multimedia content are extracted.
- original multimedia content is quantized using vector quantization techniques.
- the extracted invariant features and quantized original multimedia content are encrypted by a private key to form a digital signature.
- received multimedia content is verified through a feature-based registration process by comparing invariant features that are extracted from the digital signature of original multimedia content with the invariant features extracted from received multimedia signal. If the feature-based registration indicates that received multimedia content is significantly modified, such as when multimedia content has been forged, the authentication fails. If, however, received multimedia content is only slightly modified, and such modifications are acceptable based on a pre-defined threshold, then a refined authentication process is used to determine the integrity of original multimedia content.
- multimedia content is pre-processed to ensure better consistency of results.
- the invariant features may be extracted manually by inserting one or more landmarks into original multimedia content.
- Original multimedia content is quantized by one or more codewords, which may be labeled by corresponding index codes.
- a maximum tolerant authentication error may be defined and one or more codewords may be represented by a codebook.
- an error control coding scheme may be used to re-organize the codewords to minimize the weighted distance measure between adjacent codewords.
- a pseudo-gray code may be assigned to the reorganized one or more codewords.
- the error control coding scheme may be used to process pseudo-gray codes based on a pre-defined distance threshold and produce a result that may be hashed, to decrease a signature size and to increase security against undetected modifications.
- FIG. 1 is an illustrative diagram of different authentication levels with different robustness and accuracy.
- Fig. 2 is an illustrative diagram showing an exemplary application of the present invention.
- Fig. 3 is an illustrative diagram showing a second exemplary application of the present invention.
- Fig. 4 is a flow diagram showing a process of signing original multimedia content and a process of verifying received multimedia content.
- Fig. 5 is a flowchart showing the steps of signing original multimedia content and steps of verifying received multimedia content.
- Fig. 6 is a block diagram showing an optional Error Control Coding step in forming a multimedia signature.
- Fig. 7(a) is an illustrative diagram of adjacent codewords in a current coding space before Error Control Coding step having centers and disordered indexes.
- Fig. 7(b) is an illustrative diagram of adjacent codewords in a current coding space after an Index Assignment and re-coding to Gray Code.
- Fig. 7(c) is an illustrative diagram of adjacent codewords in a current coding space and of a new Error Control Coding space where the codewords, that are within a pre-set authentication threshold of an original multimedia input sample (block), are mapped
- Fig. 8 is a flow diagram of an invariant feature registration process.
- Fig. 9 is an illustrative diagram which shows different quantization regions of an original multimedia content according to their importance.
- Fig. 10(a) is an illustrative diagram which shows an original multimedia sample (block) S that is assigned to a closer codeword CI.
- Fig. 10(b) is an illustrative diagram which shows a received multimedia sample (block) S k that is assigned to a closer codeword C2 after some distortion during transmission.
- Fig. 11 (a-f) are illustrative diagrams showing verification results of original, forged and time compressed audio signals.
- Fig. 12 (a-c) are images of original, printed original, and printed forged Identification Cards.
- Fig. 12 (d-h) are illustrative diagrams which show verification results of the printed original and the printed forged images of ID cards.
- the same reference numerals and characters, unless otherwise stated, are used to denote like features, elements, components or portions of the illustrated embodiments.
- An authorized user 201 can login into a multimedia service center 203, read and print secure original multimedia content 205 related to its business, and later present the printed multimedia content 205 for a business purpose or otherwise.
- An authorized authenticator 207 can also login into the server 203 to verify the legitimacy of received multimedia content, and check whether the received content has been altered.
- distortions may be present that have been introduced during a formation of a secure original multimedia content or its transmission. Some examples of such distortions are scaling, color or intensity changes, slight rotation, or distortions caused by addition of noise. These types of distortions are acceptable in the verification technique provided by the present invention.
- FIG. 3 shows a multimedia commerce transaction where a customer 301 purchases a type of multimedia content (not shown), for example music, from a media distribution center 305. After the purchase, the multimedia content is distributed 302 to the customer 301. Assuming that the customer 301 has some reasonable rights to modify the bought music, such as to convert between audio formats of different players (MP3 (306), TwinNQ (307), or WAV (308) player), the consumer will necessarily need to convert from one format to another.
- the content-based signature scheme in accordance with the present invention is able to satisfy the needs of the customer while ensuring robust and flexible verification by the content owner.
- FIG. 4 depicts a content based signature scheme which has two parts ⁇ a process of forming an original multimedia content signature and a process of verifying received multimedia content.
- the first step is to detect certain invariant features (not shown) from an original multimedia content I 0 (401), that are robust to various modifications such as adding noise, scaling, and rotating.
- invariant features relating to images are comer points, average values of blocks and histograms, whereas some examples of invariant features relating to video are color values of microblocks and histograms.
- Quantizing of original multimedia content may be implemented by dividing original multimedia content into one or more source blocks, and associating the source blocks to corresponding one or more codewords (not shown) that are either pre-determined, or automatically defined during the quantization process.
- vector quantization has been extended to a more general scope to include the scalability and flexibility of multimedia content quantization.
- an operation is considered to be a quantization operation as long as the operation will scale multimedia content to some particular level based on certain pre-determined sets of rules. For example, if there are four types of objects, such as a square, a triangle, ellipse and a circle, and three types of colors, red, green and blue, there may be 12 possible objects based on two criteria - their shape and their color. If one criterion is predetermined to be more important than the other, e.g. the shape is more important than the color, then these objects may be quantized in four groups, based on their shape.
- an authorized authenticator gets a digital signature 409 and received multimedia content 411.
- Received multimedia content 411 is quantized 414 and processed to obtain the extracted invariant features (not shown).
- the digital signature 409 is decrypted by using a public key 421, and the extracted invariant features 583 and their attributes 583' of original multimedia content 401 decrypted from the received digital signature 409 are compared to the extracted invariant features 593 and their attributes 593' of received multimedia content 411. The result of comparison 419 is then evaluated.
- the codewords 506 representing original multimedia content are selected from a codebook 505 (see Fig. 5).
- the codebook is either pre-determined, in which case it is referred to as a universal codebook or an off-line codebook, or it is automatically formed during the process of quantization, in which case it is referred to as an on-line codebook.
- the universal codebook is transmitted with the digital signature and it is used in the verification process of received multimedia content.
- the on-line codebook is separately, and automatically, created on the verification side.
- An error control coding scheme and hashing may be used in cases where multimedia content is large.
- ECC error control coding
- the processes of signing and verification have certain common procedures such as invariant feature extraction, vector quantization, (optional) alignment, (optional) error control coding, and (optional) hashing.
- the verification process also has certain unique procedures such as content registration and refined authentication. Each of these procedures are discussed in more detail below.
- Invariant feature extraction The first step in either the process of signing, or that of verification, of a multimedia content is a step of extracting invariant features. Referring to Fig. 5, the flowcharts of the signing and the verification processes are depicted.
- Original (401) and received (411) multimedia contents are initially pre-processed (502) to ensure better consistency of the results.
- the typical pre-processing could be low-pass filtering, intensity normalization, etc.
- the invariant features 503 are extracted from pre-processed original (or received) multimedia content by using one or more detection methods. Different detection methods work well with different classes of multimedia content and different kinds of modifications, so it is preferred to use several detection methods in this step.
- One example of a detection method that may be used is disclosed in C.G. Harris et al., A combined corner and edge detector, 4 th Alvey Vision Conference, pp. 147-151 (1988) the contents of which are incorporated by reference herein.
- Another example of a detection method is disclosed in C. Xu, J. Wu, and Q.
- the symbol fl means an intersection of all sets of detected features to guarantee a consistent detection result.
- their invariant attributes 503' are computed in a small area around the invariant features 503. For example, if an invariant feature is a comer point of an image, then the attributes such as a position of the comer, an angle of the comer, a magnitude of the comer and a moment are computed in a neighborhood of that comer.
- the invariant attributes of the can be expressed as follows:
- A **i ' k, ' • • • ' A ⁇ kp ' (2)
- a k is a set of invariant attributes 503' associated with the invariant feature A:,. 503.
- the extraction of invariant features 503' can be implemented through the manual insertion of landmarks into the content and later extracting the landmarks from the content.
- a printed document may not contain invariant features that can be detected by the detection methods.
- certain landmarks which are subsequently detected as invariant features, may be manually inserted.
- the initial signature generation or verification step requires the detection of invariant features 503 of original (or received) multimedia content 401 (or 411), and the computation of corresponding attributes 503'.
- VQ Vector quantization
- the vector quantization encoder pairs up each source vector with the closest matching codeword, thus "quantizing" it.
- the actual encoding is then a simple process of sequentially mapping a source vector to a codeword from a codebook.
- original multimedia content 401 is partitioned into small blocks (not shown).
- a small block may be referred to as an audio sample.
- the portioning can be defined as:
- /_ [*, x 2 ... x n ) (4)
- Each block has r elements, and all blocks have the same size, i.e., the same number of elements. It may be assumed that C represents either the universal, or the on-line codebook 505 of the size K, having K codewords y(i).
- One feature of the vector quantization technique provided herein is scalability.
- This verification technique allows for breaking original multimedia content or forming a codebook with corresponding codewords according to different levels of importance attached to different regions of original multimedia content.
- the quality of reconstructed multimedia content mainly depends on the size of the codebook and the size of source blocks.
- Multimedia content may be divided into several regions, where each region has a certain level of importance. Consequently, the codebook may be adjusted to provide different number of codewords corresponding to different regions based on the importance levels attributed to such regions.
- Another unique feature of vector quantization is its flexible procedure in codebook formation.
- the maximum tolerant authentication errors can be defined on particular dimensions, particular blocks, and even on particular regions by assigning to them different thresholds. Therefore, different parts of the content can be authenticated with different levels of robustness according to the relative importance of each part.
- One example of a codebook formation is an off-line training, which is used to obtain a universal codebook so that both the procedure of signature formation and the procedure of signature verification can be simple and fast.
- Another example of the codebook formation is an on-line training, which is used to obtain a more optimal codebook so that the accuracy of authentication can be improved.
- the procedure of codebook formation is actually a clustering of given source blocks, as explained below.
- a clustering algorithm classifies these blocks into clusters by natural association according to some similarity measure, which represents the distance between a source block and the corresponding codeword.
- the clustering algorithms are chosen to maximize the degree of similarity among blocks within clusters, and to minimize the degree of similarity among blocks of different clusters. In other words, the clustering algorithms try to maximize intra- cluster distances and minimize inter-cluster distances. Further details about clustering algorithms called LBG algorithms may be found in Y. Linde, et al., An algorithm for vector quantizer design, IEEE Trans. Commun. Vol.CON-28, No.l, pp.84-95 (1980), the contents of which are incorporated by reference herein.
- the main elements of the clustering algorithm are choosing K initial cluster centers z ⁇ , z 2 , ..., Z , distributing the blocks among the K clusters using a minimum distance criterion, determining a centroid of each cluster of blocks and naming it a new cluster center, and comparing the new cluster centers with the previous ones. If the change is below a pre-set threshold, the algorithm is terminated; otherwise, steps 2-4 of the algorithm are repeated.
- the final message may be made by combining all of them in the process referred to as alignment 507.
- a signature S 409 can be formed by using the private key 408 to sign the message M , which is a result of alignment 507 and represents the invariant features A 503 and their attributes 503' combined with the VQ information W 506:
- the formed signature can either be put into the header of the multimedia file or stored in a verification center.
- the signature consists of four parts.
- the first part includes global information concerning original multimedia content.
- global information for an audio signal may include the total number of samples, the number of channels, sampling rate, a bit number of each sample, the length of quantizer, and the size of the codebook.
- Global information for an image may be the original image size, block size, codebook size, or the number of bits for each pixel.
- Global information for a video signal may include the total number of frames, the size of a frame, the codebook size, or the size of a quantization vector.
- the second part of the signature represents the invariant features that are extracted from original multimedia content. It is mainly used for feature-based registration of received multimedia content with respect to original multimedia content by comparing and matching the invariant features decrypted from the signature with those extracted from received multimedia content.
- the third part represents VQ codewords or corresponding index codes which can be used to authenticate received multimedia content, and can be either hashed or non-hashed.
- Non-hashed codewords or index codes can be used to indicate positions of modified parts in received multimedia content.
- Hashed codewords may be used to determine that received multimedia content has been modified, but they cannot be used to localize the positions of modified parts.
- the fourth part contains control information such as distance threshold which is used to define a range of allowable distortions of received multimedia content.
- the signature size may vary with an accuracy level of the verification procedures. For example, for a grey-scale (8 bit for pixel) image with a size of 256*256 pixels, if the codebook size is 256 and the block size is 8*8 pixels, the size of the first part of the signature will be around thirty-six bits. As an example, if ten invariant features are used in the second part, each point requires 50 bits to describe its feature values; thus the size of the second part is 500 bits.
- the third part is very clear: 128 bits are needed for the hashed VQ codewords and 8192 bits for the non- hashed VQ codewords.
- the resulting signature sizes are comparable to those obtained by traditional cryptography, which are usually between 1024 and 8192 bits.
- Some examples include entropy coding, re-quantizing the non-hashed codewords, coding of only the difference signs (or changes) of adjacent VQ codewords, transform VQ, and predictive VQ.
- an exemplary a content registration procedure 800 is shown.
- the invariant features 503 and 593 are used to register received multimedia content 411, since this procedure does not significantly increase the size of the signature.
- a comparison (not shown) and a matching procedure (not shown) are first done between the features decrypted 583 from the signature 409 and the features extracted 593 from the received media 411.
- This procedure referred to as a feature-based registration 513, is particularly useful where authenticating scanned content.
- the registration information (not shown) can be included in the signature 409 with a private key 408.
- a set of invariant features 593 is extracted from I w 411 using the same feature detection algorithms as in signature generation (1).
- the attributes 593' associated with the extracted invariant features are computed.
- K" S e (13)
- the K' and K" may not be the same, but there still may be some matching relationship between K' and K" .
- This matching relationship may be estimated (820) by comparing the invariant features of decrypted digital signature and received multimedia content, and determining whether received content has been slightly or significantly distorted based on a pre-defined distance threshold. If the distance between the corresponding features falls within the threshold, it means that the received content is still legitimate. Otherwise, the authentication fails.
- Attributes 593' of the extracted invariant features 593 preferably include their positions, magnitude, moments, and other attributes computed in the neighborhood of the invariant features, such as slope and central frequency for audio, or the differential Gaussian filtering for image.
- the invariant features extraction is a relative measure and has limitations. For example, the extracted attributes 503 and 593 for images remain invariant only within a scaling range from -15% to 15% of the original size.
- the content registration procedure has a hierarchical structure. The first step is a rough estimation of possible transformations of received multimedia content (820). Also, two sets of invariant features representing original (817) and received (818) multimedia contents are compared to estimate any modifications.
- the modifications are estimated by computing the matrix of moments derived from the co- ordinates of invariant features. Further details may be found in Q.B. Sun, et al., Recovering modified watermarked image with reference to original image, SPIE 3697, EI99, San Jose (1999), the contents of which are incorporated by reference herein. After this rough estimation, insensitive invariant attributes are computed and associated with the invariant features.
- the second step is a refined matching of the invariant attributes to obtain one-to-one point correspondence 822. It is important to note that all invariant attributes are in the same domain.
- the one-to-one matching 822 is done by directly looking for the best-matching point in another set based on a pre-defined threshold.
- the third step is a spatial registration used to determine whether a received content must be slightly adjusted, such as scaled or rotated or shifted to adjust the extracted invariant features of received multimedia content with respect to the decrypted invariant features.
- the fourth step registers the changes of magnitude, if there are any, to modify the intensity of the extracted invariant features of received multimedia content with respect to the decrypted invariant features.
- the registration is finalized by solving surface spline functions among all matched point-pairs (solving «+3 linear equations if the number of matched pairs is n).
- the rough authentication result can be given as a percentage representing the total number of matched features.
- the next step in the verification process is a refined authentication 516.
- the result of the refined authentication process 516 is "true.” This is similar to the procedure of traditional verification. However, if the received message has been changed, the refined authentication result of the present invention yields a distance measure 526. It is up to a particular application to determine, by selecting a proper threshold T, whether received multimedia content 411 has been altered, and to detect which parts of media have been modified.
- the consistence between the VQ codewords 596 quantized from the received media 411 and the set of codewords 586 decrypted from the signature 409 is verified.
- the vector quantization of received media 411 is conducted in one of two ways based on the results from the feature-based authentication 513. If the modifications are not severe, the codebook 555 can be generated on the verification side, if an on-line codebook is used. The codebook must remain the same on both the signing and the verification sides to ensure that a consistent and a reliable authentication is obtained.
- the codewords 506 can be hashed 508 and put into the signature 409 to verify the consistency between the codebooks 505 and 555 on both sides.
- a codebook with 256 codewords and 16 bytes per codeword needs only 4Kbyte, which can be transmitted within seconds.
- TTP Trusted Third Party
- the size of the codebook 505 will not impose much burden on the system.
- a universal codebook may be used. In such cases, the codebook is transmitted in advance.
- the refined authentication can be conducted in two ways: if the VQ codes were not hashed, received multimedia content 41 1 can be re-constmcted 525 according to the codewords decrypted 586, de-compressed 524 and decoded from the signature 409 and the codebook 555. The error distance measure 526 between the reconstructed media 525 and the received media 411 is then computed directly. In this case, the modified part of received multimedia content 411 can be localized. If the VQ codewords were hashed 508, quantized codewords 596 representing received multimedia content 41 1 based on the codebook 555 are obtained, hashed and compared to the hashed codewords 586 decrypted from the signature 409 bit by bit.
- An optional step of hashing 496 represents a processing of the VQ codewords in a signature to yield a hashing result that uniquely represents original multimedia content. Even the slightest distortion of it produces different hashing results. This is why hashing is done in conjunction with an error control code scheme, which ensures that slightly distorted multimedia content still yields the same hashing result.
- an alternative embodiment having an e ⁇ or control codes (“ECC") procedure of forming the multimedia signature 405 may be used.
- the ECC procedure requires re- organizing of codewords and assigning of pseudo-gray code labels to re-organized codewords, so that the Hamming distance between adjacent pseudo-gray codes equals 1.
- the VQ codewords of received multimedia content may not equal the VQ codewords of original multimedia content, but their pesudo-gray code labels will still be within the Hamming distance. This means that the codewords, representing the acceptably distorted parts of received multimedia content will be located very close to the codewords representing source blocks of original multimedia content. This would allow for a possible error location during a process of authentication.
- ECC error control coding
- FIG. 6 an alternative verification technique which incorporates an error control code procedure 600 is illustrated.
- original multimedia content 401 or received multimedia content 4111 is vector quantized to obtain a set of codewords 625.
- the codewords are then re-organized in the index assignment 626 (see Fig. 7(a)) so that the distance between adjacent codewords is minimized.
- pseudo-gray code labeling 627 of the re-organized codebook is obtained.
- the resulting codes are ECC decoded 629, and finally hashed 636 to obtain a secure hashed code 640 that is insensitive to acceptable distortions of received multimedia content.
- a set of codewords after VQ Coding 625 is illustrated.
- the indexes 701 of the adjacent codewords 702 are disordered, namely, they are CW n , CW b , and CW-.
- the centers of codewords 703 are illustrated.
- the whole codebook 505 (or 555) is re-organized under the criterion that a minimum distance between adjacent codewords is minimized.
- the re-organization can be implemented by optimization algorithms. For example, the binary switch algorithms introduced by the Zeger reference or by P. Knagenhjelm, Hadamard Transform-a Tool for Index Assignment, IEEE Trans. IT, Vol. 42, No. 4, pp. 1139-1151, the contents of which are incorporated by reference herein.
- a diagram representing a set of codewords 702 after re-organization 626 is illustrated.
- the indexes 701 of the adjacent codewords are in the following order: and Cr ⁇ i -
- the indexes are re-coded 704 to obtain pseudo-gray code labels 627: GC b _ ⁇ , GC b , and GC b +i.
- a property of pseudo-gray code is that the Hamming distance between adjacent code labels representing adjacent codewords is only 1. Therefore, even if a distorted block of received multimedia content is assigned to a different codeword from the corresponding source block of original multimedia content, since the distance between those two codewords is very small, they will be close to each other.
- the received index code labels representing the codewords of received multimedia content will not match the corresponding original index code labels 701 representing the codewords of original multimedia content. Nevertheless, the received index code labels will be located close to the original index code labels 701. In other words, the hamming distance between the received index code labels and the original index code labels 701 is 1.
- the original index code label 701 may be 11100111
- the new index code label may be 11101111.
- the positions of the resulting pseudo-gray codes can be classified into perfect positions and possible error positions. A perfect position means that a digit is at the correct position. A possible error position means that the digit may be at the wrong position and that it should be evaluated.
- the index code label is then ECC encoded and decoded by using an ECC decoding scheme 629.
- ECC decoding scheme 629 The detailed ECC encoding technique can be found in L. A. Bassalygo, et al., Coding or Partially Localized Errors which is incorporated by reference herein.
- an ECC decoded set 629 of codewords 702 is illustrated.
- the source block is shown as black square 715, and the pre-determined authentication threshold is depicted by a radius of a circle 732.
- the codewords 735 inside the circle will be taken into account for ECC.
- the codewords that are within a pre-specified threshold are mapped onto a new code New ⁇ V 729 shown in a new ECC coding space 730.
- the ECC decoded set 629 of codewords is insensitive to accepted modifications, such as scaling etc. Therefore, a hashed code 636, insensitive to acceptable modifications, can be obtained.
- the received index code labels will not be located in the threshold circle of the corresponding original index code labels, so a different mapping onto a new code NewW ⁇ 729 will result.
- a different new code, which is a result of ECC decoding is hashed, and a different hashing result 636 is produced. Therefore, acceptable modifications may be distinguished from malicious manipulations.
- multimedia content having regions with different levels of importance is illustrated.
- the VQ codebook 505 can be changed so that the important image regions within the image have stronger protection.
- regions 901 and 902 which may correspond to a face and hands, are more important than a region 903.
- more codewords 506 can be assigned to source vector blocks corresponding to the important regions 901 and 902, while less codewords 506 can be assigned to vector blocks corresponding to less important region 903.
- the maximum e ⁇ or on dimensions as well as a maximum block average error in the procedure of forming the codebook 505 is limited.
- the maximum error on dimensions represents the maximum of errors on dimensions with respect to each element of the original and the received block.
- a maximum average block error is an average value of errors for each element.
- the unequal protection technique may be important for application files such as MPEG-4 that have different security requirements associated with different regions.
- National Map authorities publishes architectural design maps. The maps have many layers of plots representing, for example, plumbing, or electrical wiring. Such layers are very important and must not be altered, whereas other layers representing interior or exterior design are not as important and some modification is allowed. As a result, different levels of importance are attached to different layers, and they are coded differently, thus allowing for greater protection of layers such as plumbing or electrical wiring.
- the unequal protection technique may be used in optical character recognition ("OCR") applications where it is necessary to distinguish between similar alphanumeric symbols. For example, the letter "1' and the number "1 " in Roman Fonts may have the same codeword.
- similar letters may be assigned to one codeword to obtain a consistent verification result.
- the similar letters may be distinguished by increasing the resolution of files to be authenticated. If in some applications the security is a primary concern, these ambiguous patterns may further be distinguished either by increasing the resolution of these special parts or by adding some special codewords to take care of these special cases.
- FIGs. 10(a) and (b) depict exemplary multimedia content which is partitioned into small blocks of pre-determined size, one of which is a source block S k 1001 before it is signed.
- a codebook (not shown), having the codewords CI 1002 and C2 1003, is also pre-determined.
- the distance between the codeword C, 1002 and the codeword C 2 1003 may be regarded as small under a pre- determined codebook and block sizes.
- a source block S ⁇ . 1001 which is located slightly closer to C, (1002) than to C 2 (1003), is assigned to the codeword C, (1002).
- the source block S k 1001 is authenticated, the distances from S k 1001 to C, and C 2 may slightly change due to certain distortions in the transmission procedure.
- the source block S k 1001 is now assigned to C 2 1003, and a wrong authentication result is produced.
- This issue is addressed by setting a particular threshold, dependent upon a particular codebook format and a particular codeword formation scheme. If there are p codewords (not shown) whose distances to S ⁇ are below a threshold T , these p codewords are treated as virtually equivalent with respect to S k . Referring to Figs.
- FIGs. 12(a-h) depict an original (1201), a printed original (1202), and a forged (1203) ID cards.
- the printed original image 1202, and the forged image 1203 are authenticated using the previously discussed procedures, and the two forged parts 1205 and 1206 have been detected. (See Figs. 12(f), (g), and (h)).
Abstract
Description
Claims
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US10/149,685 US20040128511A1 (en) | 2000-12-20 | 2000-12-20 | Methods and systems for generating multimedia signature |
AU39673/01A AU3967301A (en) | 1999-12-20 | 2000-12-20 | Methods and systems for generating multimedia signature |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US17271999P | 1999-12-20 | 1999-12-20 | |
US60/172,719 | 1999-12-20 | ||
US17730000P | 2000-01-21 | 2000-01-21 | |
US60/177,300 | 2000-01-21 |
Publications (2)
Publication Number | Publication Date |
---|---|
WO2001047278A2 true WO2001047278A2 (en) | 2001-06-28 |
WO2001047278A3 WO2001047278A3 (en) | 2002-03-14 |
Family
ID=26868386
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2000/034803 WO2001047278A2 (en) | 1999-12-20 | 2000-12-20 | Methods and systems for generating multimedia signature |
Country Status (2)
Country | Link |
---|---|
AU (1) | AU3967301A (en) |
WO (1) | WO2001047278A2 (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2003055222A2 (en) * | 2001-12-21 | 2003-07-03 | Koninklijke Philips Electronics N.V. | Increasing integrity of watermarks using robust features |
WO2005081476A1 (en) * | 2004-02-19 | 2005-09-01 | Openmind Networks Limited | A content management system |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5987459A (en) * | 1996-03-15 | 1999-11-16 | Regents Of The University Of Minnesota | Image and document management system for content-based retrieval |
-
2000
- 2000-12-20 AU AU39673/01A patent/AU3967301A/en not_active Abandoned
- 2000-12-20 WO PCT/US2000/034803 patent/WO2001047278A2/en active Application Filing
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5987459A (en) * | 1996-03-15 | 1999-11-16 | Regents Of The University Of Minnesota | Image and document management system for content-based retrieval |
Non-Patent Citations (4)
Title |
---|
KLIOT M ET AL: "INVARIANT-BASED SHAPE RETRIEVAL IN PICTORIAL DATABASES" COMPUTER VISION AND IMAGE UNDERSTANDING, ACADEMIC PRESS, US, vol. 71, no. 2, 1 August 1998 (1998-08-01), pages 182-197, XP000766982 ISSN: 1077-3142 * |
LIN C-Y ET AL: "ISSUES AND SOLUTIONS FOR AUTHENTICATING MPEG VIDEO" PROCEEDINGS OF THE SPIE, SPIE, BELLINGHAM, VA, US, vol. 3657, 25 January 1999 (1999-01-25), pages 54-65, XP000949137 * |
SCHMID C ET AL: "LOCAL GRAYVALUE INVARIANTS FOR IMAGE RETRIEVAL" IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, IEEE INC. NEW YORK, US, vol. 19, no. 5, 1 May 1997 (1997-05-01), pages 530-535, XP000693746 ISSN: 0162-8828 * |
SCHNEIDER M ET AL: "ROBUST CONTENT BASED DIGITAL SIGNATURE FOR IMAGE AUTHENTICATION" PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP) LAUSANNE, SEPT. 16 - 19, 1996, NEW YORK, IEEE, US, vol. 3, 16 September 1996 (1996-09-16), pages 227-230, XP002090178 ISBN: 0-7803-3259-8 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2003055222A2 (en) * | 2001-12-21 | 2003-07-03 | Koninklijke Philips Electronics N.V. | Increasing integrity of watermarks using robust features |
WO2003055222A3 (en) * | 2001-12-21 | 2003-11-06 | Koninkl Philips Electronics Nv | Increasing integrity of watermarks using robust features |
WO2005081476A1 (en) * | 2004-02-19 | 2005-09-01 | Openmind Networks Limited | A content management system |
Also Published As
Publication number | Publication date |
---|---|
WO2001047278A3 (en) | 2002-03-14 |
AU3967301A (en) | 2001-07-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20040128511A1 (en) | Methods and systems for generating multimedia signature | |
Tkachenko et al. | Two-level QR code for private message sharing and document authentication | |
US6879703B2 (en) | Method and apparatus for watermarking images | |
US6834344B1 (en) | Semi-fragile watermarks | |
RU2606056C2 (en) | Documents protection and authentication method and device | |
Holliman et al. | Counterfeiting attacks on oblivious block-wise independent invisible watermarking schemes | |
Yahya | Steganography techniques for digital images | |
KR100465950B1 (en) | Information processing apparatus and its control method, storage medium storing computer program | |
CA2109554C (en) | Secure document and method and apparatus for producing and authenticating same | |
US7028902B2 (en) | Barcode having enhanced visual quality and systems and methods thereof | |
US7656559B2 (en) | System and method for generating a signed hardcopy document and authentication thereof | |
US8175322B2 (en) | Method of digital watermark and the corresponding device, and digital camera which can append watermark | |
Kailasanathan et al. | Image authentication surviving acceptable modifications using statistical measures and k-mean segmentation | |
US20040039914A1 (en) | Layered security in digital watermarking | |
JP2003527778A (en) | Protection of the legitimacy of electronic documents and their printed copies | |
CA2304611A1 (en) | Method for generating and verifying digital watermarks and for exchanging data containing digital watermarks | |
Chandramouli et al. | Digital watermarking | |
JP2003510925A (en) | Authentication of digital data products using signatures and watermarks | |
CN110503592B (en) | Two-dimensional code anti-counterfeiting method and system based on digital watermark | |
CN108491913B (en) | Method for authenticating QR (quick response) code with secret beautification function | |
Dittmann et al. | Invertible authentication for 3D meshes | |
US8285995B2 (en) | System and method for digitally marking a file with a removable mark | |
WO2001047278A2 (en) | Methods and systems for generating multimedia signature | |
US6792535B1 (en) | System and method for digitally marking a file with a removable mark | |
Tran et al. | Lsb data hiding in digital media: a survey |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AK | Designated states |
Kind code of ref document: A2 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CR CU CZ DE DK DM DZ EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NO NZ PL PT RO RU SD SE SG SI SK SL TJ TM TR TT TZ UA UG US UZ VN YU ZA ZW |
|
AL | Designated countries for regional patents |
Kind code of ref document: A2 Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE TR BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
DFPE | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101) | ||
AK | Designated states |
Kind code of ref document: A3 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CR CU CZ DE DK DM DZ EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NO NZ PL PT RO RU SD SE SG SI SK SL TJ TM TR TT TZ UA UG US UZ VN YU ZA ZW |
|
AL | Designated countries for regional patents |
Kind code of ref document: A3 Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE TR BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG |
|
REG | Reference to national code |
Ref country code: DE Ref legal event code: 8642 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 10149685 Country of ref document: US |
|
122 | Ep: pct application non-entry in european phase | ||
NENP | Non-entry into the national phase in: |
Ref country code: JP |