WO2004084418A2 - Method of providing a covert security feature - Google Patents

Abstract

A method of providing a covert security feature on an article. The method comprises generating data defining indicia to be provided on a substrate of the article, the indicia comprising a plurality of elements, wherein the data defines a security feature constituted by a characteristic of the elements which varies in a predetermined manner. Providing the indicia on a substrate of the article using a marking material and in accordance with the data, wherein the providing step causes the characteristics of the elements to vary so that the security feature cannot be determined from a visual inspection of the indicia on the substrate but the presence of the security feature can be deduced following a statistical analysis of the variation in the said characteristic of the elements of the indicia on the substrate using the predetermined manner in which the security feature was defined.

Description

METHOD OF PROVIDING A COVERT SECURITY FEATURE

The invention relates to methods for providing a covert security feature on an article such as a document of value. The presence of such a feature can be used to determine the authenticity of the article and/or to prevent a copying process being undertaken.

Photocopiers, DTP systems, scanners/PCs/printers, digital press and other image reproduction devices are becoming more sophisticated as manufacturers attempt to enable much more accurate or realistic reproductions to be generated. However, there are certain types of document, for example documents of value, where reproduction is undesirable. There have therefore been many attempts in the past to prevent reproduction of such documents. These have included incorporating into the documents certain security features such as optically variable devices and the like. In addition to this, equipment manufacturers have devised methods which enable a photocopier or DTP system to recognise that it has been presented with a document of value which should not be reproduced and then stop the reproduction process. Typically, this has involved incorporating certain codes such as digital watermarks and the like into documents. The photocopier looks for the code or watermark and a decision is made as to whether or not to copy dependent upon the presence or absence of the code or watermark.

Examples of prior art describing these approaches include W000/7356, 099/53428, EP917113, DE19800316, US5949903, EP675631, EP946039, EP940780, EP585724, EP382549, EP440142, EP529746, US5367577, EP522769, EP485694, EP067898, EP342060, EP961239, and EP490457.

For example, EP-A-961239 describes an approach in which machine readable data is imperceptibly embedded into visual data by altering the distribution of the ink slightly to encode the data. The data can then be retrieved and decoded using a suitable detection device such as a scanner; the scanner may form part of a copying apparatus .

The problem with these known approaches is the need to detect and read the embedded code which is undesirable. In accordance with a first aspect of the present invention, a method of providing a covert security feature on an article comprises generating data defining indicia to be provided on a substrate of the article, the indicia comprising a plurality of elements, wherein the data defines a security feature constituted by a characteristic of the elements which varies in a predetermined manner; and providing the indicia on a substrate of the article using a marking material and in accordance with the data, wherein the providing step causes the characteristics of the elements to vary so that the security feature cannot be determined from a visual inspection of the indicia on the substrate but the presence of the security feature can be deduced following a statistical analysis of the variation in the said characteristic of the elements of the indicia on the substrate using the predetermined manner in which the security feature was defined.

In contrast to the known approaches in which it is important that the code remains detectable and decodeable following printing on a substrate, the invention recognises and takes advantage of the fact that the step of providing indicia on the substrate will degrade the elements of the indicia to such an extent that the precise form of the original feature cannot be derived from the indicia. The invention goes on to recognise, however, that the existence of the feature (even though its precise form cannot be determined) can be deduced by analysing the statistical properties of the elements in the knowledge of the predetermined manner in which the feature was defined.

The characteristics of the elements are typically selected to have statistical properties that are strongly related to the quality of the image on the substrate, and/or to reflect the empirically determined properties of the distribution of marking material on the substrate. For example, in a design which consisted of closely arranged line segments the characteristic might be the tendency of segments to merge. The measurements of such a characteristic would have statistical properties that were a well defined function of the marking ink and the substrate .

Thus, in the case of banknotes, the present invention measures properties of putative banknotes which are directly related to their quality of production and hence provides an indicator of likely authenticity.

Typically, the predetermined manner comprises varying the data defining the characteristic of known elements in a known manner between a plurality of values, although it is possible to allow a subset of the elements to have characteristics that take on values that are not predetermined to act as a control set or give a bench mark.

Conveniently, the data defining the characteristic is varied between two values although three or more values are possible.

The characteristic itself typically comprises the width or length of line segments or dots defining the elements but could comprise instead or additionally colour

(hue, chroma or lightness) of the elements. In the case of line segments, these may be physically spaced apart or different parts of a continuous line, e.g. a spiral.

The principle of the invention, therefore, covers any feature which cannot be reproduced with absolute precision by the printing and scanning processes or by photocopying processes using electrostatic means, for instance. In the former category are any geometrical features such as line width, line length, regularity of location of featur-es. In the latter are any colour features using subtle colour changes that cannot accurately be reproduced and possible metamerism to confuse the reproduction processes.

In one embodiment, the modified pattern will be composed of two or three elements that are repeated in a regular pattern to form a design, the appearance being the result of the location and orientation of the elements. Alternatively the design may comprise continuous lines as in a medallion pattern. This can be contrasted with "digital watermarking" where each pixel of an image is described by a multibit parameter and where the placing of any elements is non predictable and indeed conveys the information which is the essence of the image.

The essence of the preferred implementation is the printing onto a substrate of a pattern, with measurable characteristics that are very sensitive to the scanning and printing operations that form a necessary part of the unauthorised copying process. The pattern to outward appearances is typically a conventional line pattern but in reality is made up of sets of near identical elements distributed in an ostensibly random manner according to a sequence known only to the originator. The distinctions between the elements, although clearly defined in the original filmwork and printing plates or in electronic form for digital printing, are obfuscated by the printing process with the inevitable indeterminancy arising from the unpredictable transfer and flow of ink onto a substrate. At the edges of the print. The authenticity of the substrate, e.g. a banknote, is guaranteed by the detection of the sequence following the scanning of the banknote and application of an algorithmic detection process based on a statistic analysis of the data.

An important feature of the invention is the fact that the originating pattern is not revealed to those who have access only to the printed copy. The unpredictable spreading of ink brought about by the printing process ensures that all that is available is a distribution that has statistical properties deriving from the original file. A consequential feature of the invention is the requirement to characterise the operation of the printer and the ink flow on the document so as to predict the optimal difference in dimensions of the elements forming the pattern. Thus, on a fibrous substrate, the unpredictability of ink flow and the "dot gain" or ink spread may result in lines, whose original widths were identical, differing by perhaps ten per cent on the printed document, i.e. there is a variation in width along the length of a line.

Preferably, the statistical analysis comprises determining for each group of known elements corresponding to each of the plurality of values, a distribution of the actual values of the characteristic obtained from the indicia on the article, and comparing the distributions, whereby if the difference between the distributions is greater than a threshold, the presence of the security feature is deduced. It will be appreciated that other statistical analyses could be carried out as discussed later.

In order to enhance the covert nature of the security feature, preferably the predetermined manner utilizes a pseudo-random sequence. The advantage of a pseudo-random sequence is that although it is very difficult to deduce from the resultant feature, it can be regenerated by repeating the same algorithm which will normally respond to a seed or seeds .

This leads to a particularly advantageous aspect of the invention in which the seed(s) is incorporated into or onto the substrate. This means that the precise nature of the security feature does not need to be known by the system carrying out the method. The system only needs to be provided with the algorithm for generating the pseudo- random sequence and it obtains the seed(s) for the particular feature on the substrate from the article itself. This leads to the possibility of individually coding articles such as documents of value including digitally printed secure documents. In such documents of value it is common to include a serial number and the seed(s) could be obtained from one or more of the digits of the serial number. Preferably, however, use is made of other features of value such as security line patterns, name of country of issue, name of currency etc as will be explained in more detail below. As a further alternative, both the seed and indicia could be defined by the same graphical pattern.

In other examples, other information, typically numeric such as an account number, on a substrate could be used to provide the seed(s) . The seed number could be derived from any random source, typically a hardware device, or from such data as the time and date of printing. However, the seed has to be present in the document be it as a printed number or as some pattern of colours or lines (e.g. bar code) .

Alternatives to pseudo-random sequences include any function of a seed number such as a one way hash of a serial number. The significance of a one way hash is that the original number cannot be deduced from the hash and so no recoverable information is added.

An important further feature of the invention can be obtained when the variation in the characteristic of the elements is chosen such that after the article provided with the indicia is copied, the presence of the security feature cannot be deduced following the said statistical analysis . This enables copies of articles to be detected on the basis that a genuine article should have retained the security feature in a deducible form while a counterfeit copy, due to degradation introduced by the copying process, will no longer allow the security feature to be detected. An advantage of this is that copiers do not need to be provided with special detection equipment and can simply allow copies to be made but those copies can be detected as counterfeit when presented to a recipient .

As mentioned above, the invention can be applied to a wide variety of articles but is particularly suitable for a document of value such as a banknote, travellers cheque and the like. In accordance with a second aspect of the present invention, a method of inspecting indicia on an article to determine if the indicia includes a security feature provided in accordance with the first aspect of the present invention comprises scanning the indicia to generate data defining the said characteristic of elements of the indicia; analysing the data in accordance with the predetermined manner associated with the security feature to determine the statistical distribution of the variation in the said characteristic; and indicating that the security feature is present if the statistical distribution satisfies predetermined conditions.

The scanning step may be carried out using a flatbed scanner, or a web or digital camera or any other suitable scanning or image capture device.

As mentioned above, the presence or absence of the security feature can be utilized in different ways depending upon the equipment involved in carrying out the method. In the case of a photocopier or desk top publishing system, the security feature can be treated as an anti -copy feature whose presence could be used to prevent a copy being made, cause only a partial copy to be issued, cause the copy to be damaged or marked, or indicate an illegal action is occurring. The feature can also be used as a teller assist feature to distinguish between a genuine document and a counterfeit .

In the case of document of value handling apparatus, the presence or absence of the security feature can be used as a test of authenticity. For example, in the case of a banknote acceptor, if it is determined that the security feature is not present on a proffered banknote, the banknote can be immediately returned to the user as unacceptable . An example of methods and apparatus according to the invention will now be described with reference to the accompanying drawings, in which:- Figure 1A illustrates part of a line pattern on an enlarged scale;

Figure IB illustrates the Figure 1A pattern after modification; Figure 1C illustrates the Figure IB pattern after printing;

Figure ID illustrates the Figure 1C pattern after further copying;

Figure 2 shows part of the pattern in Figure IB, further enlarged;

Figures 3A and 3B illustrate graphically the distribution of line segment thicknesses for the Figure 1C and Figure ID example respectively;

Figure 4 is a block diagram of apparatus for preparing filmwork of the security feature;

Figure 5 is a block diagram of a teller assist arrangement; and,

Figure 6 is a block diagram of a photocopier.

In preferred examples of the invention, indicia defining line structure patterns commonly used in banknote printing, for example "medallion" are used. These are designs which consist entirely of lines. The lines are generally printed in a single "spot" colour (as opposed to being generated by a four or six colour printing process) , although they could also be rainbowed in accordance with methods used by high security printers .

Design and Embedding Process

Initially, an electronic file is constructed conveniently in vector format, where line segments are described by their end points.

The modifying pattern is conveniently generated by the use of pseudo random numbers, that is to say, numbers whose values are fairly evenly distributed and which lack any regular pattern that would be discernible by human perception unaided by analytical tools.

(a) Random Number Generation A very simple example would be numbers generated by the two initial values 73 and 37 according to the rule

R(n+1) = ( R(n) x 73 + 37 )mod(256)

Mod (256) indicating that the remainder should be taken after dividing by 256.

If R(0) = 23, for example, we would find that

R(l) = (23 x 73 + 37)mod(256) = 1,716 mod(256) = 180 since 1, 716 = 256 x 6 + 180

R(2) = (180 x 73 + 37)mod(256) = 13 , 177mod (256) = 121 since 13,177=256x51 +121

and so on. The sequence of random numbers generated would depend upon the 3 values, 73,37 and 23.

There are much more sophisticated ways of producing such sequences, well described in standard literature, the illustration serving only to show that a roughly random set can be produced by a few initial numbers or seeding values.

There are many different ways in which seed values could be generated and recorded. In the case of banknotes, a preferred approach is to make use of a security line pattern on the banknote. In one example, a security line pattern consists of a set of regular hexagons, each hexagon containing 8 parallel lines orientated horizontally, vertically or at +/- 45 degrees. (See Figure 1A) .

The orientation of the lines within a hexagon is used to obtain a seed value, i.e. each orientation is assigned an arbitrary value:

Vertical lines = 10

+45 degrees = 20

Horizontal lines = 30 -45 degrees = 40

Orientation of lines at three or more particular points relative to the edge of the banknote is determined. The corresponding numbers are then noted and then used in a formula such as that described above in order to generate a set of random numbers .

In another approach the name of country/currency is used as it appears on the banknote : each letter corresponds to a number e.g. A=10, B=ll, C=12 etc. The first three letters of country/currency name are chosen and corresponding numbers are noted. E.g. Dollar: D=13 , 0=24, L=21. In another embodiment the random numbers are subjected to a one way "hash" before being used as determinants of the pattern. That is to say the numbers are transformed into a new set of numbers by a rule that cannot be reversed. This means that it is not possible from the generated pattern to deduce the seed values and thus the pattern is not the holder of any information.

Another possibility is the use of hardware to produce random numbers. There are several mechanisms for this based on the inevitable "noise" that exists in electrical circuits.

The set of random numbers, once generated, is used to select the elements to be modified in the manner described below.

(b) Pattern Construction

An example of a suitable pattern is shown in Figure 1A consisting of a set of regular hexagons, each hexagon containing 8 parallel lines, the lines being horizontal, vertical or diagonal. The pattern is then divided up in a regular manner that can be described by a simple algorithm, (the division being conceptual rather than consisting of visible physical lines) . The simplest method is to divide the pattern into a set of congruent rectangles although a shape that is more adaptive to the original pattern can improve the statistical properties of the pattern modification. Figure 2 shows an example of a single such hexagon. The subdividing rectangles are outlined with narrower lines.

Typically the line widths will be roughly 50-250μm and the line separation roughly 50-270/xm. The dimensions are appropriate to particular printing conditions as indicated below.

The rectangles are then labelled in an orderly manner, typically in a rasterised form from left to right and from top to bottom. The order is immaterial provided the decoder uses the same order as the encoder.

In its simplest form the process divides the rectangles into two sets, distinguished by the fact that the width of lines in one set is greater than the widths in the other set . The sets are chosen by use of the random numbers. Each rectangle in order is associated with a random number, the numbers taken in sequence. A criterion is established which determines whether any given random number shall correspond to a wide or a narrow line. For example, if the sum of digits of the given random number is odd then the line should be wide whereas if the sum is even the line should be narrow. If a rectangle is designated as wide then the segment of line passing through that rectangle is widened by maybe 4 pixels. A decision has to be made about lines that are partly in two rectangles, and again the deciding criterion is irrelevant providing it is shared with the decoding software.

Figure IB shows the modified form of the Figure 1A pattern while Figure 2 shows a single hexagon from Figure IB. The same process can be applied to other patterns. It may be that instead of rectangles, rhombi are used as the subdividing shape in order to fit the pattern more closely. This would lessen the number of cases that occur in the previous pattern where the rectangle includes intersections of two or more hexagons and hence gives less reliable information. The resultant modified pattern is stored as data in electronic form.

The modified pattern is conveniently converted into a raster format file for film plotting purposes and a printing plate produced.

A typical banknote is printed on a fibrous cotton or non-cotton based substrate where the precise spread of ink is randomised to a degree by the paper's fibrous nature.

It could also be printed on a polymer based substrate. The precise nature of the printing and the substrate is important in any application because it governs the choice of line width and the modifications described above. Under these conditions the widths of some lines in the wide set would overlap with those in the narrow set due to the random spread. It would not be possible from any single sample of a banknote to determine the line widths for any particular line segment in the original electronic file.

Figure IC illustrates the result of printing the pattern on paper. Inspection Process

In order to determine whether or not a banknote is genuine under the present scheme it is first of all necessary to obtain a digital image of appropriate quality for analysis. Using the range of parameters indicated above a suitable digital image may be obtained by using a cheap flat bed scanner at a resolution of around 300 dpi. The fact that this is less than the printing resolution does not invalidate the analysis because of the statistical nature of the decoding process.

In some circumstances an instant image retrieval might be required rather than the slow machinations of a scanner. This is possible by using a web camera or a digital camera or image capture device in a suitable environment. This environment would include lighting of sufficient intensity, accurate placing of the optical instrument so as to achieve little distortion, or distortion of known characteristics which can be corrected by appropriate geometrical transformation .

The first process in recovering the data is to establish the precise location and orientation of the virtual grid on the image of the banknote being examined. This is achievable in a number of ways. Knowledge of the position of the pattern relative to some prominent feature on the banknote offers a simple method. Alternatively the pattern itself may have certain macro features which provide fiducial points capable of establishing orientation. Thus in a pattern of hexagons there may be points at which three hexagons all having vertical lines intersect. Determination of a pair of such points may be sufficient to establish a representative direction and also provide a scaling factor to offset distortions due to paper stretching, inaccuracy of scanner lead screw or any similar mechanical defect.

After such a process the image, in the case of the Figure IC pattern, can be subdivided into rectangles again as illustrated in Figure 2, except that there will be a degree of blurring of the image following the printing and scanning processes.

The second process is to generate the appropriate set of random numbers, the same set that was used to modify the original file. The requirement here is to find the seeding values. If these have been based on the serial number of the banknote this number will probably be obtained by using optical character recognition on the scan of the serial number. If instead the number generation is based on the preferred approach of using the configuration of the hexagon pattern then the image will be analysed to ascertain, for example, the orientation of lines in a set of hexagons, this being a very simple coding device.

The set of virtual rectangles can then be associated with the set of random numbers in the same order that was used in the encoding process . The result is that the image is conceptually subdivided into rectangles, each rectangle being classified as corresponding to a wide line or a narrow line. (There may be a greater number of classifications in other embodiments.)

The data within each rectangle is then analysed to produce a metric corresponding to the width of the line. If the data is bilevel (as in Figure 2) , that is to say the scan values are either black or white, the width of the lines is assessed by a simple counting process. There will generally be corrections to allow for random pixel values using knowledge of the type of configurations allowed in the original pattern.

In some embodiments the data will be contone in form. That is to say, each pixel will have attributed a value typically in the range 0 to 255 according to how dark or light it is.

In this case when assessing the width of a line a threshold has to be determined to decide whether a particular pixel is deemed to be part of a line or part of the interstitial space. Adjustments have to be made for the configuration of pixels allowing for the fact that closely bunched pixels tend to produce generalised darker areas on account of ink flow in the printing process or signal spreading in the scanning process. It should also be borne in mind that the orientation and alignment of the rectangles can never be exact because, in addition to the above mentioned distortions there is a variable stretching of the substrate, mechanical inaccuracies in scanner lead screws etc.

The result of these processes is that each set of rectangles has an associated set of numbers corresponding to the width of lines within the rectangular cells.

Figure 3A illustrates this situation in the case where printing takes place on a substrate which is a coated paper producing little "dot gain" or ink spread as shown in Figure IC. Scan resolution is 300 dpi and line width is digitized into 256 values (from 0 for all black to 255 for pure white, i.e. zero thickness) . Here the line thickness is assessed essentially by totalling the values for each pixel within a rectangle. It can be seen that the thickness values for the rectangles containing the supposedly narrow lines vary from roughly 8 to 23 with a peak at roughly 14 (These thickness values are in arbitrary units) . The values for the other rectangles containing supposedly thick lines range from roughly 9 to 25 with a peak at 17.

In the original electronic file the values for the rectangles corresponding to narrow lines would have all been very close to 14, giving a spike rather than the bell shape. Similarly the values for the rectangles corresponding to wider lines would give a spike around the value 17 in the original file. Thus it can be seen that by measuring the width of a line in a rectangle, so great is the overlap of values that for the majority of values it would not be apparent whether the measurement came from a designated narrow or designated wide rectangle. The separation of the two distributions is an indication of the presence of the embedded pattern but there are two points to resolve if this is to be a reliable method. The first is the question as to whether by chance a similar separation of distributions might be achieved. The second is whether some copying process could reproduce the distributions sufficiently accurately to render uncertain any decision as to whether the banknote is an original or copy.

One way to answer this question is to calculate the mean of the distributions and consider how that may vary.

As an example suppose that 100 of each type of rectangle were sampled from a banknote and the means of the two distributions calculated. The question arises as to how much the mean varies on account of the chance variation from one note to another. In the example the mean of the narrow rectangles might be 14 and of the wider rectangles might be 17 and it is important to know over what sort of range these numbers might vary if we took a large sample of notes .

From the distribution indicated on the graph (Figure 3A) the standard deviation of line widths is approximately 2 for both the wide and narrow lines. This spread can be taken under reasonable assumptions as an estimate for the spread of the distributions that would be found in any banknote that is sampled. If we take a sample of 100 rectangles and calculate the mean, this mean value will vary less than the individual samples as one intuitively expects. Using elementary statistics the standard deviation of the means will be reduced from the standard deviation of the original individual rectangle distribution by a factor equal to the square root of 100, i.e. 10, giving the value 0.2. If we keep sampling banknotes at random the probability of a mean being more than 2 standard deviations from the mean of the means is very small according to statistical theory for the likely distributions. Thus we would find that the means of the banknote samples lay between 14 -0.4 and 14 + 0.4. for the narrow rectangles and 17 -0.4 and 17 + 0.4 for the wide rectangles . The means of any samples would clearly be distinct. A measure of the confidence of the separation can be calculated by dividing the difference of the means by their standard deviation. This gives an empirical discrimination factor that depends upon the particular environment, printer, scanner, substrate.

Analysis of Counterfeit Copy

If a copy is made from a banknote the scan of the banknote will include the spread of data already described together with distortions resulting from the scan itself. If the image is printed there will be the same or more degradation than arose from the original printing. For example ink jet printing causes greater degradation than offset printing. The result of copying and printing the Figure IC design on an inkjet printer or copier is shown in Figure ID and analysed in Figure 3B. The result is that typically the standard deviation of the distribution is doubled. In the case above there would still be separation of the means of the distributions but the factor that was calculated by comparing the mean separation with the standard deviation would be halved. Measurement of this factor would pick out a counterfeit banknote. Figure 3B shows the result of a similar analysis of a copy of the original on which Figure 3A is based.

The statistical analysis can be made in a more sophisticated manner but to do so would require careful modelling of data for every printer and scanner involved in the process and with only a small gain in accuracy over the simple assumptions of roughly normal distribution etc. In practice an empirical evaluation of the range of parameters in original and copied banknotes will provide a means for automatic discrimination.

In the case that the embedded pattern made adjustment to the lengths of lines rather than the widths the analysis would be more or less identical to that above. There would, however, tend to be skewing of the distribution because both the printing process and the scanning process tend to elide features which are not clearly separated, as would be the case with line patterns which are fragmented according to some random rule. In the case where the embedded pattern uses variation in colour it would be possible to use two or three "spot" colours which differed very slightly and where the differences were blurred by the differential absorption and flow of ink on the substrate. A scanner with a suitably selected filter would be the instrument for image retrieval .

Filmwork Preparation

An example of apparatus for preparing filmwork including a pattern of the type described above modified to include an anti-copy or security feature, is shown in Figure 4. The apparatus comprises a processor 10 coupled to a pair of data stores 11,12. The data store 11 stores data defining the colour content of pixels of an image such as an image of a banknote and including a line pattern of, for example, the Figure 1 type described above. The processor is also connected to a random number generator (RNG) 13. In operation, the processor 10 provides seed(s) to the RNG 13 which then generates a pseudo-random sequence which the processor uses to modify data from the store 11 defining the Figure 1 pattern so as to change line segment widths between thin and thick values, the modified pattern then being stored in the store 12.

The processor 10 then supplies the pattern in the store 12 to a conventional imagesetter 14 which exposes the banknote pattern onto a suitable record medium such as a film in a conventional manner. From this film, a printing plate is formed which is then used to print the pattern on to banknote paper. As explained above, the printing process will cause the thick and thin widths as originally defined by the data in the store 12 to be further modified or degraded such that a number of line segments intended to be "thin" appear "thick" and vice versa. (In this context, a typical variation in thickness between "thick" and "thin" lines is 5%.) In addition, a pattern or other indicia from which the seed(s) can be derived may also be printed e.g. a serial number. Teller Assist Embodiment

Figure 5 illustrates in schematic, block diagram form a teller assist arrangement for use in banks, retail outlets etc. The device comprises a banknote transport system 50 of conventional construction driven by a motor 52. Banknotes are supplied singly by a teller to an input slot 54 and are then transported by the transport system past at least one detector 56. The detector 56 allows a processor 58 to generate digital data defining the colour content of each pixel of the banknote which is stored in a similar manner to that described above in connection with

Figure 5. Again, in a similar manner, the processor 58 analyses the stored data utilizing a random number generator 60 to determine whether or not a security feature is present on the document. As described above, this may involve obtaining one or more seeds from the scanned image which are used by the RNG 60 to define a pseudo-random sequence used by the processor 58 to determine whether each region in a line pattern should be "thick" or "thin".

Following the analysis, the processor 58 causes a message to be displayed on a monitor 62 indicating whether the banknote is genuine or not. Depending upon the structure of the transport system 50, the result of the determination can then be used in a variety of ways to control further handling of the banknote.

In a simple approach, whatever the outcome, the processor 58 controls the motor 52 to reverse the transport systems so that the banknote is fed back out through the slot 54 for further handling by the teller.

In another approach, the transport system 50 may be connected with a store (not shown) , the processor 58 allowing the motor 52 to continue operating so that genuine banknotes are fed on for storage in the store. On the other hand, if the banknote is determined not to be genuine then the motor 52 is reversed and the banknote fed back out through the slot 54. In this case, it would not be necessary to display the outcome of the test on the monitor 62 since the performance of the transport system 50 would inevitably provide that information to the teller.

In a further option, the teller could control the motor 52 via an input device 64 such as a keyboard, mouse or touch screen thus allowing the teller to determine how the banknote is further handled i.e. returned to the teller or send to a store.

Photocopier Embodiment

Figure 6 illustrates in block diagram form a typical photocopier structure. The apparatus of Figure 6 comprises a processor 20 coupled to a RNG 21 and to an input scanner

22. In use, a document to be copied is supplied to the scanner 22 which will create digital data defining the colour content of each pixel of the scanned document, the digital data being stored in a store 23. The processor 20 then analyses the data in the store 23 in the manner described above to determine whether or not the anti -copy, security feature is present in the original document. This analysis may involve obtaining one or more seeds from the scanned image, for example a serial number or the like, which are used by the RNG 21 to define the pseudo-random sequence used by the processor 20 to determine whether each region in the line pattern should be "thick" or "thin". Alternatively, the random number sequence could be prestored. Following the analysis, the processor 20 can then make a decision as to whether or not the anti-copy feature is present on the original document. If it is then it is likely that the copier will wish to prevent the document being copied. In one application, therefore, a processor 20 could cause an output device or printer 24 to cease operation and the data could be deleted from the store 23 to prevent further reproduction. In another option, the processor 20 could control the output device 24 to generate a poor copy of the original document which would be easily recognisable as such while in a further alternative, a special marking could be applied by the output device 24 on the copy indicating that it is a copy and not an original.

Claims

1. A method of providing a covert security feature on an article, the method comprising generating data defining indicia to be provided on a substrate of the article, the indicia comprising a plurality of elements, wherein the data defines a security feature constituted by a characteristic of the elements which varies in a predetermined manner; and providing the indicia on a substrate of the article using a marking material and in accordance with the data, wherein the providing step causes the characteristics of the elements to vary so that the security feature cannot be determined from a visual inspection of the indicia on the substrate but the presence of the security feature can be deduced following a statistical analysis of the variation in the said characteristic of the elements of the indicia on the substrate using the predetermined manner in which the security feature was defined.

2. A method according to claim 1, wherein the predetermined manner comprises varying the data defining the characteristic of known elements in a known manner between a plurality of values.

3. A method according to claim 2, wherein the data defining the characteristic is varied between two values.

4. A method according to claim 2 or claim 3, wherein the statistical analysis comprises determining for each group of known elements corresponding to each of the plurality of values, a distribution of the actual values of the characteristic obtained from the indicia on the substrate, and comparing the distributions, using a statistical method, whereby if the difference between the distributions is greater than a threshold, the presence of the security feature is deduced.

5. A method according to claim 4 whereby the statistical method involves determining a confidence level for the separation of the means of the distributions by dividing the difference of the means by their standard deviation.

6. A method according to any of the preceding claims, wherein the predetermined manner utilizes a pseudo-random sequence .

7. A method according to claim 6, when dependent on any of claims 2 to 5 , wherein the value of the data defining the characteristic of successive elements is chosen in accordance with the pseudo-random sequence.

8. A method according to claim 6 or claim 7, wherein the pseudo-random sequence is generated from a seed or seeds, the method comprising providing further indicia on the substrate defining the seed(s) .

9. A method according to claim 8, wherein the seed is defined by all or part of a serial number on the substrate, or a one-way hash function of a serial number.

10. A method according to any of claims 1 to 8 , wherein the seed is defined by a characteristic of a pattern on the substrate.

11. A method according to any of the preceding claims, wherein the indicia define a graphical pattern of which the security feature forms a part .

12. A method according to claim 10 and claim 11, wherein the seed and indicia are defined by the same graphical pattern.

13. A method according to any of the preceding claims, wherein the variation in the characteristic of the elements is chosen such that after the article carrying the indicia is copied, the presence of the security feature cannot be deduced following the said statistical analysis.

14. A method according to any of the preceding claims, wherein the elements comprise spaced or continuous line segments .

15. A method according to claim 14, wherein the characteristic which is varied comprises the width and/or length of the line segments.

16. A method according to any of the preceding claims, wherein the characteristic of the elements which is varied comprises the colour of the elements.

17. A method according to any of the preceding claims, wherein the indicia are printed on the substrate.

18. A method according to any of the preceding claims, wherein the article comprises a document of value such as a banknote, travellers cheque, financial instrument, personal identification document such as a passport, driving or vehicle licence, ticket, voucher, postage or fiscal stamp, security label, certificate of authenticity, security packaging, brand label or confidential document.

19. A method of inspecting indicia on an article to determine if the indicia include a security feature provided in accordance with a method according to any of the preceding claims, the method comprising scanning the indicia to generate data defining the said characteristic of elements of the indicia; analysing the data in accordance with the predetermined manner associated with the security feature to determine the statistical distribution of the variation in the said characteristic; and indicating that the security feature is present if the statistical distribution satisfies predetermined conditions .

20. A method according to claim 19, when dependent on claims 8 to 10, further comprising obtaining the seed(s) from the article and thereafter generating a pseudo-random sequence using a known algorithm and based on the seed(s) for use in analysing the data.

21. A method according to claim 19 or claim 20, wherein the scanning step is carried out using a flatbed scanner, a web or digital camera, or image capture device.

22. A method according to any of claims 19 to 21, when dependent on claim 4 or claim 5, wherein the predetermined conditions are satisfied if the difference between the distributions is greater than a threshold.

23. A method of copying documents in which the document is inspected in accordance with a method according to any of claims 19 to 22, the copying process being terminated and/or a partial or full copy of the document being generated with a suitable marking, if the security feature is deduced to be present on the original document.

24. Document handling apparatus for processing documents, the apparatus including an inspection system for carrying out a method according to any of claims 19 to 22, and a control system for handling a document in a predetermined manner if the security feature is deduced to be present.

25. Apparatus according to claim 24, the apparatus comprising a copier which is adapted to terminate the copying process and/or to mark any partial or complete copies produced with a suitable marking, if the security feature is present on the original .

26. Apparatus according to claim 24 for handling banknotes, the apparatus treating as genuine a document found to include the security feature.

27. Apparatus according to claim 24, the apparatus being adapted to return a document found not to include the security feature, to a user.