GB2548546A - Method and apparatus for detection of document tampering - Google Patents

Abstract

The invention relates to a computer-implemented method, and apparatus, of testing a digital image (10a, 10b) of a cheque (check) or other document for indications that it was subject to tampering. The digital image comprises a set of pixels each having one or more pixel values. The method comprises analysing the digital image to identify contiguous sets of pixels over a predetermined size in which all the pixels in a set have pixel values which match those of the other pixels in the set to within a predetermined margin. The identification of image regions may comprise testing pixels within a cell and counting the number of pixels whose pixel values match a reference pixel. Further disclosed is a computer-implemented method and apparatus for testing a document for features suggestive of tampering which includes the identification of marked pixels, defining a set of tiles, calculating, for each of the tiles, a tile image density and analysing tile image densities to determine whether any is suggestive of tampering. The inventions may have particular use in electronic cheque clearing systems and allow the detection of washing, coating or other alteration of the cheque or document which results in a background pattern being removed.

Description

METHOD AND APPARATUS FOR DETECTION OF DOCUMENT TAMPERING

The present invention is concerned with detection of tampering with a document. In particular it is concerned with detection of tampering with cheques.

The word "cheque", or in US English "check", refers to a document authorising a bank to make a payment from a specified account to a specified person or entity. Despite the increasing use of purely electronic means of payment, paper cheques are still used for a large number of transactions. Typically, to obtain payment the recipient of the cheque (the "payee") deposits it at a first bank (the "collecting bank"). Through the clearing system, the cheque is transferred from the collecting bank to a second bank on which the cheque was drawn (the "drawer's bank"). The second bank then makes the specified transfer of funds to the collecting bank, and the funds are credited to the payee's account.

Traditionally the clearing system has been based on transfer of the paper cheque itself from the collecting bank via the clearing system to the drawer'. Given the large daily volume of cheques handled by the clearing system, and despite the use of automatic sorting methods, this exchange of paper documents is a substantial burden to the banking system.

An alternative to reliance on exchange of the paper cheques themselves is to electronically scan the cheques, so that banks exchange digital cheque images and data rather than the paper documents. Making use of modern computer networks, this process can be considerably less onerous than sorting and transport of the paper cheques. Cheque clearing based on exchange of electronic images has at the time of writing been adopted in some countries but has yet to be taken up in others. In particular the UK clearing system is currently reliant on exchange of paper cheques, (although some images are currently exchanged, but not to the exclusion of the exchange of paper) and it is planned that it will move entirely to use of scanned cheque images in the near future, the paper cheque being discarded once the image has been scanned.

One issue to be addressed in making this transition is security against fraud. Fraudsters are known to tamper with cheques in a variety of ways in order to try to obtain improper payments from banks. This may for example involve theft of bona fide cheques which are then altered prior to presentation to a bank. Tampering with a cheque covers both forgery of the cheque itself and illicit alteration of a cheque's information content. There are various ways in which cheques can be altered for fraudulent purposes. For example, by altering the payee on a completed cheque, the fraudster may divert payment into its own account. By altering the amount, the fraudster may obtain a larger payment than it is entitled to. Alteration of the cheque may be carried out in a range of different ways. For example some form of washing or coating may be applied to the cheque to remove or conceal written or printed markings before replacing them with the fraudulent information.

Various techniques have been applied to paper cheques over the years to enable tampering to be detected. Some of these relate to the cheque's paper substrate, which may be of a specific type which is difficult for a forger to replicate or obtain, and may be sensitive to chemicals used for washing or other types of tampering. Other anti-tampering techniques use ink technology, such as bleeding inks that show up attempts at tampering and magnetic inks to be read by a magnetic ink character reader (MICR). Watermarks, holograms or microprints can provide further security against forgery and counterfeit. A background marking, often in the form of a pattern, applied to the cheque - especially in the regions which are to be filled in by the drawer, such as the field for the amount payable - helps to reveal where markings have been removed by techniques such as washing, since the marking will be absent in the areas affected. The pattern is often formed by an ultra violet ink which is not apparent to the naked eye but which can be seen under ultra violet illumination or by means of a UV sensitive camera.

Some of these fraud prevention measures can be put to use only where access is available to the paper cheque itself. In a clearing system reliant entirely on exchange of digitised cheque images, opportunities for successful cheque tampering are thus potentially increased.

There is thus a need for an effective, computer-implemented means of analysing an image of a document - and especially but not exclusively a cheque - to provide an indication of possible tampering.

According to the present invention there is a computer-implemented method of testing a digital image of a cheque for indications that the cheque was subject to tampering according to any of the appended independent method claims.

According to a further aspect of the present invention there is an apparatus for testing a document for tampering according to any of the appended independent product claims.

Specific embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

Figure la shows an optical image of a cheque obtained by scanning the cheque using an imaging device sensitive to light frequencies in the visible part of the spectrum (hereinafter a "visible light image") and Figure lb is a bi-tonal version of the same image;

Figure 2 shows an inverted image of the same cheque obtained using a camera sensitive to ultra violet light (hereinafter, a "UV image");

Figure 3 shows a portion of the Figure 2 image to an enlarged scale;

Figures 4a to 4c represent, at a pixel level, cells used for image analysis in an embodiment of the present invention; and

Figure 5 is a highly schematic representation of a computing environment suitable for implementing the present invention.

Features of a cheque

The cheque 10, 10a, 10b illustrated in Figures 1 and 2 is in itself of a conventional type. It comprises a thin rectangular substrate which is paper in this example although it could in principle comprise some other suitable material, such as a polymer substrate, although not in the UK where the substrate is strictly specified and controlled. The cheque's front face carries various markings, some of them printed onto the cheque during its manufacture and others filled in by the drawer (the person completing the cheque).

The printed markings applied during manufacture include data applied during initial printing of the cheque: a bank name/logo 12, a bank branch sort code 14, a number line comprising the cheque's serial number 16, the bank branch sort code 18 and the drawer's bank account number 20. In this example these are printed in magnetic ink, in a font which is suited to reading by MICR, the identity 22 of the drawer. This is an example of a corporate cheque where the identity is often included. A personal cheque will not show this.

Additionally the pre-printed parts of the cheque delineate fields for insertion of information by the drawer. These fields comprise: lines 24 for the payee and 26 for the legal amount - the amount payable, in words, a date line 28, a box 30 for the courtesy amount (in figures) and a signature space 32.

The UV image 10b of Figure 2 includes all of these markings but additionally shows a background marking 34. This is not seen in the visible light image 10a, nor upon inspection of the physical cheque with the naked eye, because it is formed by what is sometimes referred to as invisible UV ink - ink that is not visible to, or at least difficult to discern with, the naked eye, but is revealed by some other means. In this case the background marking 34 is visible when the cheque is imaged using a photodetector sensitive to light in the ultra violet frequency range. The background marking may take a variety of forms but typically has, in relation to the other markings on the cheque, a relatively high spatial frequency. This makes its presence or absence on any chosen part of the cheque easier to discern. In the illustrated example the background marking 34 is a pattern of wavy lines closely spaced from one another and overlapping one another, which provides a high spatial frequency of variation from light to dark. In the illustrated example the background marking 34 covers much of the area of the cheque, including almost all of the information-bearing area save for the number line 16, 18, 20. In other cases the background marking may be more selectively applied.

Imaging the cheque

In a cheque scanning and clearing system in which the present invention is employed, the physical cheque is typically presented by the payee to a collecting bank (or other institution). The collecting bank scans the physical cheque to obtain at least one digital image.

In practice, two digital images may be obtained.

The first image is the visible light image 10 of Figure la and is obtained using a camera sensitive to light in the visible part of the spectrum. In Figure lb the same optically sampled image is shown in bitonal form - that is, pixels are represented as either black or white, rather than in RGB format as in Figure la.

The second image is the UV image 10b of Figure 2 and is obtained using a camera sensitive to light in the ultra violet part of the spectrum. This image has been inverted - that is, each pixel colour value has been subtracted from the maximum colour value. In the present example pixels are RGB encoded and the maximum value is 255, so that a colour value of 5 becomes 255-5 = 250 upon inversion. Hence for example black becomes white, and vice versa.

Scanning devices capable of scanning cheques to provide both visible light and UV images are commercially available and will be familiar to the skilled person. Typical scanners have a motorised feeder which moves the cheque past an imaging head incorporating both visible and UV imaging devices. The scanner may additionally read the code line by MICR. The visible light and UV images 10a, 10b, are then output in digital form to a computer for processing.

Note that the term "image" is used herein to refer both to the displayed image - e.g. on the screen of a computer - and to the digital data set by which this image is represented.

First method of image analysis - overview

The computer processing of the images includes steps intended to detect cheque tampering. That is, the images are subject to a computer-implemented test intended to detect whether the physical cheque was tampered with prior to scanning. The test is not necessarily expected to be 100% accurate.

It may be subject to false positives, so that cheques which fail the test and are thus considered suspect will need to be subject to further enquiry or examination before appropriate action is taken by the relevant institution. In the normal course of events the images will be digitally transmitted from the collecting bank to the drawer's bank through a computer network, so in principle the image analysis may be carried out anywhere in the computer network - by the collecting bank, the drawer's bank, or by some intermediate authority. The present invention is adaptable to any of these possibilities.

The method used to analyse the images comprises the following steps:

Step 1. The bi-tonal version 10a of the optical image is analysed to identify informationbearing fields.

Step 2. The information-bearing fields are each divided into tiles.

Step 3. Each tile is scanned to identify aberrant cells within the tile. If aberrant cells are identified, these are listed to form an output for further analysis.

These steps will now be described in more detail.

Image analysis step 1

The information-bearing fields which are used in the present embodiment are indicated by rectangles in Figure 2 and comprise a payee field 24a containing the payee line 24, a legal amount field 26a containing the legal amount lines 26, a date field 28a containing the date line 28, a courtesy amount field 30a containing the written amount box 30, and a number line field 16a containing the cheque's serial number 16, the bank branch sort code 18 and the drawer's account number 20. This selection of fields is not the only one that is possible, within the scope of the present invention. One could for example treat the entire area of the cheque bearing the background pattern 34 as a single field. The fields shown are rectangular but other shapes could be used.

It might be imagined that the locations of the relevant items - the amount lines, signature space etc. - on the cheque would be standardised, so that the fields could be defined at predetermined locations on the cheque which would not change from one cheque to another. But in fact the layout and proportions of cheques varies somewhat. To accommodate this, in the present embodiment the positions and/or the size and/or the proportions of one or more of the fields 16a, 24a, 26a, 28a, 30a are determined on the basis of analysis of one of the cheque images. More specifically in the present embodiment this analysis involves the use of character recognition. This is a technique well-known to the skilled person. It is often referred to as "OCR", for "optical character recognition", although in this case it could in principle be applied to either the visible light image 10 its bi-tonal derivative 10a or the UV image 10b. The image is analysed to find a chosen character, or chosen characters, associated with the specific field. To the left of the courtesy amount box 30 for example there is typically a character denoting the currency of the payment, such as "£" or "$". The analysis carried out in the present embodiment includes a search for this currency character and the written amount field 30a is defined at a location to the right of it. Likewise to the left of the payee line 24, the words "PAY" are typically shown on the cheque. The analysis includes a search for this character string, and the payee field 24a is defined at a location to the right of it.

The fields 16a, 24a, 26a, 28a, 30a in this embodiment all contain information relating to the transaction being carried out by means of the cheque - the cheque's date, the payee etc. Hence they are likely areas for tampering. They are also, in this example, all filled with the background marking 34. That is, the area of the fields is entirely taken up either with the background marking 34 or with other printed matter of the cheque.

Image analysis step 2

The fields 16a, 24a, 26a, 28a, 30a are divided into tiles. Figure 2 shows a single tile 36 within the payee field 24a. In this example each of the fields is broken down into a set of square tiles such as the example shown. The tiles 36 cover the entire field area without overlapping. This manner of division of the fields into tiles is not the only one possible within the scope of the present invention. Any other suitable pattern of tiles and/or shape of tiles could be adopted.

Image analysis step 3

The scanning of a tile to identify aberrant cells will now be described with particular reference to Figures 3 and 4.

In the present embodiment the scanning process is carried out on the UV image 10b represented in Figure 2, since this includes the background marking 34. In relation to documents of other types. however, e.g. those having some suitable background marking which is visible in a visible light image, the technique could equally well be applied to the visible light image.

The UV image 10b consists of a set of pixels. In Figure 3 a mesh 42 is superimposed on the area of the image forming the tile 36. Each square within this mesh is a single pixel. In the present embodiment the pixel values are represented according to the RGB colour model. Hence each pixel is represented by three pixel values. Colour models are very familiar to the skilled person and the principles need not be explained here. Other models for representing a pixel are known and may be adopted in other embodiments of the invention. In particular, the invention could be implemented using digital images in which pixels are represented by fewer than three values, e.g. a greyscale image in which each pixel is represented by a single value.

The scanning process involves testing of what will be referred to as cells within the tile 36. A cell is a group of pixels. In the present embodiment the pixels making up a cell are contiguous. More specifically, in the present embodiment every cell is a square block of pixels having a width of five pixels along the horizontal ("X") direction and a depth of five pixels along the vertical ("Y") direction. Other sizes or shapes of cells may be adopted in other embodiments. The size of the cell is chosen with reference to the expected spatial frequency of the background marking 34. Figure 3 shows example cells 44a, 44b.

Each cell can be identified by reference to a coordinate, say the coordinate of its top left pixel. The cell 44a in Figure 3 is thus cell (0,0), the origin of the coordinate system being at the top left of the tile 36.

Each cell is subject to an aberrance test. According to the outcome of this test, the cell is identified either as aberrant or as normal. Cells identified as aberrant are suggestive of tampering and their coordinates are stored for further reference. In the present embodiment the aberrance test has only two possible outcomes - normal or aberrant. The aberrance test as applied to the cell 44a proceeds as follows.

One pixel in the cell serves as a reference pixel. In the present embodiment the reference pixel for each cell is its top left pixel. For the cell 44a, which is cell (0,0), the reference pixel is thus pixel (0,0). However any other pixel in the cell could be chosen as the reference pixel in another embodiment.

The aberrance test comprises a preliminary test and a full test. The full test is carried out only if the result of the preliminary test shows it to be necessary. In the present embodiment the preliminary test comprises a comparison of the reference pixel against another pixel in the cell, to establish whether they are similar to within a chosen margin. In the present embodiment this comparison is made using the R,G,B pixel values and the chosen margin is plus or minus five in each of the pixel values. Also the reference pixel is compared to its immediate neighbour in the same row.

Consider the example represented in Table 1 below. Each box represents a pixel and contains its RGB values. The reference pixel in the top left has pixel values R=10, G=10, B=10. In the preliminary test the pixel immediately to the right of the reference pixel is compared against the reference pixel. If it has R value between 5 and 15; G value between 5 and 15; and B value between 5 and 15 then it is taken to match the reference pixel and the full test is to be carried out on the cell. If not, the cell is treated as non-aberrant and the process move on to the next cell (see below). In the example, the values do fall within the relevant ranges, so the full test proceeds.

Table 1

Where the full test is carried out on the cell, this involves comparing each of the remaining pixels in the cell against the reference pixel in the same manner and counting the number that match the reference pixel. At a programming level, this may for example be done using a pair of nested loops to scan along each row or column in succession. The full test thus returns a value which may in this example be between 1 and 24. A high value shows that the cell contains a large number of matches, which is considered indicative of potential tampering. Cells whose matching pixel count is above a chosen threshold are identified as aberrant and their coordinates are recorded to form the output of the process. For example cells may be treated as aberrant if their matching pixel count is above 90% of the number of pixels in the cell (more than 21, in the present example).

The cells may in principle be tested in any order but in the present embodiment this is done in a raster pattern, with cells on the first row being tested in sequence, then cells on the second row being tested in sequence and so on. To clarify the point, Figure 4a shows the first cell 44a to be tested which is cell (0,0). Figure 4b shows the second cell to be tested which is cell (1,0) and which overlaps the first. Scanning proceeds along the first row in the direction shown by an arrow in Figure 4b until the end of the row is reached whereupon the scan, the scan reverts to the first cell (0,1) of the second row (Figure 4c) and so on and passes along the second row, before moving to the third row and so on.

Scanning of the tile returns, in the present embodiment, a set of coordinate pairs for cells identified as aberrant.

Looking again at Table 1, matching pixels are indicated by a tick and non-matching pixels by a cross. The second pixel tested (third pixel, top row) has RGB values of 20,20,20 and so does not match the reference pixel and is not counted. In fact only one further pixel matches the reference pixel. This is the pixel immediately beneath the reference pixel. So the total number of matching pixels counted is two and the cell is not identified as aberrant. Note that although in this example the R,G and B values are all the same in each pixel, this is not the norm - more often they would differ.

Table 2 below gives an example of a cell which is identified as aberrant. In this case all pixels in the cell match the reference pixel to within the chosen tolerance.

Table 2

Table 3 is an example where the full test on the cell does not proceed. The reference pixel values of 10,10,10 are compared against the values of the first neighbouring pixel - 20,20,20 - and since these fall outside the relevant range there is no match and the cell is not tested further.

Table 3

The process is repeated for all of the tiles in the UV image 10b.

Interpretation of results

The treatment of the results is in part a matter for the institution carrying out the image analysis. One possibility is that if even a single aberrant cell is identified in the image then the cheque in question is identified as possibly having been subject to tampering and is referred for inspection by a human operative.

Another possibility is that results of the above described image analysis method are taken into account along with results of some other image analysis method - either by a human operative or by means of a software-implemented algorithm - to determine whether a document should be treated as potentially having been subject to tampering. A second method of analysing the digital images to provide an indication of possible tampering will now be described. The method may be additional or alternative to the first.

Second method of image analysis - overview

According to a second aspect of the present invention, the image densities of selected regions of the cheque are compared against a reference image density for the cheque. Differences between image density of one or more of the selected regions and the reference image density are - if outside a certain tolerance band - interpreted as suggestive of tampering.

Image density "Image density" is a numerical value or values attributed to a certain region of the image and calculated from the values of pixels within that region. It can be representative of the average brightness and/or colour of the image over the chosen region. In the case of an image in bitmap form (that is, an image in which each pixel has only two possible values, black or white) it can be the ratio of black pixels to white pixels over the chosen region. In a greyscale image it can be the average pixel value - that is, the sum of all pixel values divided by the number of pixels. In an image in RGB form it can for example be an average of the three values over all pixels, to provide a single image density, or the R, G and B values may each be averaged separately to give three values.

In the present method, image densities are calculated using pixel values from the UV image 10b. However the regions used for calculation of image density omit those regions of the cheque which carry markings (printing or writing) that are included in the visible light image 10a - i.e. regions bearing visible markings. This is achieved by making use of both the UV image 10b and the bi-tonal version 10a of the visible light image 10. The bi-tonal visible light image 10a is analysed to determine which of its pixels falls in an area carrying visible printing or writing. The UV image 10b is registered with the visible light image. The pixels in the UV image 10b corresponding to the pixels in the visible light image identified as carrying visible printing or writing are excluded from the calculation of image density.

To illustrate this consider the cheque represented in Figures 1 and 2. Visible markings on the cheque such as the characters and lines printed on it are easily discernible in the visible light image 10a of course and the relevant pixels can be selected by application of a suitable numerical criterion to the pixel RGB values. For instance, noting that RGB values of 0,0,0 denote perfect black, the relevant criterion might be that a pixel is treated as bearing a visible marking if its RGB values are all below some threshold. Using this criterion, the pixels of the visible light image 10a can be divided into (a) pixels corresponding to regions of the cheque which do not carry visible markings ("marked pixels") and (b) pixels corresponding to regions of the cheque which do carry visible markings.

It is found in practice - using commercially available scanning devices - that the UV and visible light images 10a, 10b are not precisely aligned. For example, the position on the physical cheque represented by pixel (0,0) in the visible light image 10a may be offset by some small distance from the position on the physical cheque represented by pixel (0,0) in the UV image 10b. Registration of the visible light and UV images 10a, 10b involves compensating for this offset. In the simplest case, this simply involves addition of some numerical offset to the pixel coordinates of one of the images.

Following registration it can be assumed that pixels in the visible light and UV images 10a, 10b having the same coordinates correspond to the same position on the physical cheque. Pixels of the UV image 10b having coordinates which match the marked pixels of the visible light image 10a can then be excluded from the calculation of the image density.

The image density thus obtained characterises what may be referred to as the background part of the cheque - the area of the cheque which lacks visible markings, although it typically does carry the background marking 34.

Reference image density

The reference image density is intended to represent the average or typical image density of the relevant areas of the cheque. It may be calculated based on a large area of the cheque. In Figure 2, a large rectangle 38 taking in the majority of the cheque's information bearing area including the date line 38, lines 24 and 26 for the payee and the legal amount, the box 30 for the written amount and the signature space, is used for this calculation.

Selection of regions for testing

The process involves selecting further, normally smaller, regions of the cheque, calculating the image density of each of the selected regions, and comparing the image density of each selected region with the reference image density.

The individual regions of the cheque selected for this process will be referred to herein as "tiles". In the present embodiment the process of selecting them proceeds somewhat similarly to steps 1 and 2 of the first testing method described above. The image is first analysed to identify information bearing fields of the cheque, e.g. using OCR as described above. The information bearing fields are then broken down into tiles for testing.

Testing

If one looks again at tile 36 in Figure 2, it will be apparent that the blot 45 affects the tile image density. In the area taken up by the blot 45, the background marking 34 is absent, so that the image density of the tile as a whole can be expected to be higher than the reference image density (following the convention that brighter pixels have higher numerical values).

In the present embodiment testing each tile involves calculating its image density as described above and then determining whether the tile's image density value differs from the reference image density by more than a predetermined margin. If it does, this is interpreted as indicative of possible tampering.

Some cheques have somewhat variable printing quality in relation to the background marking 34. For example cheques sometimes have bands where the background marking 34 is especially feint. Such variations are expected to affect the image density in those areas and may thus lead to false positives. For this reason the present method may advantageously be combined with some other mode of testing (e.g. the first method described above) or with other means of enquiry into the legitimacy of the transaction effected by the cheque.

Computer System

The methods described above can in principle be carried out in any form of computing system and it is not intended to limit the scope of the invention to any particular form of computer, network or scanner, but by way of example Figure 5 shows in block diagram form major elements of a system that can be used to carry out the invention, comprising a digital cheque scanner 50 arranged to pass data to a computing device 52 through a network 54 which may be a local area network. The computing device 52 is arranged to output data through a further network 56, which may be the internet and which may lead to a cheque clearing system through which cheque images are exchanged between financial institutions. Another computing device 58 may receive images and/or other data through the wide area network. The methods described above may be implemented in either of the computing devices 52, 58.

Claims (34)

1. A computer-implemented method of testing a digital image of a cheque for indications that the cheque may have been subject to tampering, the digital image comprising a set of pixels each having one or more pixel values and the method comprising analysing the digital image to identify image regions having more than an expected number of pixels having pixel values which match to within a predetermined margin.

2. A method as claimed in claim 1 in which the said identification of image regions comprises testing pixels within a cell to make a count of pixels in the cell whose pixel values match to within the predetermined margin.

3. A method as claimed in claim 2 in which one pixel in the cell serves as a reference pixel and the said count comprises testing other pixels in the cell to determine whether their pixel values match those of the reference pixel to within the predetermined margin.

4. A method as claimed in claim 2 or claim 3 in which a preliminary test is made by comparing two pixels in the cell and if the tested pixels do not match to within the predetermined margin then the said count of pixels is not made.

5. A method as claimed in any of claims 2 to 4 in which a cell is identified as aberrant if the count exceeds a predetermined value.

6. A method as claimed in claim 2 in which a cell is a contiguous group of pixels.

7. A method as claimed in claim 3 in which a cell is a block of pixels.

8. A method as claimed in any of claims 2 to 7 in which a cell is a block of pixels having a width of between 3 and fifty pixels and a depth of between three and fifty pixels.

9. A method as claimed in any of claims 2 to 7 in which a cell is a block of pixels having a width of between 4 and 10 pixels and a depth of between 4 and 10 pixels.

10. A method as claimed in any of claims 2 to 9 comprising testing multiple cells at different locations in the image.

11. A method as claimed in claim 10 in which the cells that are tested overlap one another.

12. A method as claimed in claim 10 or claim 11 in which each cell tested is separated from another neighbouring cell by a predetermined number of pixels.

13. A method as claimed in claim 12 in which the predetermined number of pixels is one.

14. A method as claimed in claim 1 in which the digital image is of a cheque that comprises a background pattern which appears in the digital image.

15. A method as claimed in claim 2 in which the background pattern has a high spatial frequency in comparison to other markings on the document.

16. A method as claimed in any preceding claim further comprising analysing the digital image to identify information-bearing fields of the cheque.

17. A method as claimed in claim 16 in which the analysis to identify the information-bearing fields comprises locating symbols or characters associated with the information-bearing fields.

18. A method as claimed in claim 17 in which the locating of the said symbols or characters comprises optical character recognition.

19. A method as claimed in claim 17 or claim 18 in which one or more of the information-bearing fields is broken down into multiple tiles and the method of testing of any of claims 1 to 15 is applied to each of the resultant tiles.

20. A method as claimed in any of claims 1 to 19 in which the document has background markings which are detectable by an optical sensor sensitive to light in the ultra violet part of the spectrum and in which the digital image is obtained using such an optical sensor.

21. A computer program comprising instructions for carrying out the method of any of claims 1 to 20.

22. An apparatus for testing a document for features suggestive of tampering, the apparatus comprising a scanner for scanning the document to provide a digital image of it, the digital image comprising a set of pixels each having one or more pixel values, and a processing device for analysing the digital image to identify image regions having more than an expected number of pixels having pixel values which match to within a predetermined margin.

23. An apparatus as claimed in claim 22 in which the scanner is sensitive to light in the ultra violet region of the spectrum.

24. A computer-implemented method of testing a first digital image representing a cheque for features suggestive of tampering, the first digital image comprising a set of pixels each having one or more pixel values and the method comprising: identifying marked pixels, which are pixels in the first digital image corresponding to regions of the cheque bearing visible markings, other pixels being unmarked pixels, defining a set of tiles, each tile being a contiguous block of pixels within the first digital image, calculating, for each of the tiles, a tile image density, the calculation being made using the unmarked pixels in the tile, any marked pixels being omitted from the calculation, analysing the tile image densities to determine whether any is suggestive of tampering.

25. A computer-implemented method as claimed in claim 24 which further comprises obtaining a reference image density and in which the analysis of the tile image densities comprises comparing each to the reference image density.

26. A computer-implemented method as claimed in claim 25 in which a tile whose tile image density differs from the reference image density by more than a certain margin is identified as aberrant.

27. A computer-implemented method as claimed in claim 25 or claim 26 in which the reference image density is the image density of a region of the cheque whose area is larger than that of any of the selected tiles, the calculation of the reference image density being made using the unmarked pixels in the said region, any marked pixels in the said region being omitted from the calculation.

28. A computer-implemented method as claimed in any of claims 24 to 27 in which the first digital image is obtained using a UV sensitive scanner, a second digital image is obtained using a scanner sensitive to visible light, and the identification of marked pixels in the first image is carried out by reference to the second image.

29. A computer-implemented method as claimed in claim 28 further comprising registering the first and second images.

30. A computer program comprising instructions for carrying out the method of any of claims 24 to 29.

31. An apparatus for testing a document for features suggestive of tampering, the apparatus comprising a scanner for scanning the document to provide a digital image of it, the digital image comprising a set of pixels each having one or more pixel values, and a processing configured to identify marked pixels, which are pixels in the first digital image corresponding to regions of the cheque bearing visible markings, other pixels being unmarked pixels, define a set of tiles, each tile, calculate, for each of a set of tiles, each tile being a contiguous block of pixels within the first digital image, a tile image density, the calculation being made using the unmarked pixels in the tile, any marked pixels being omitted from the calculation, analyse the tile image densities to determine whether any is suggestive of tampering.

32. An apparatus as claimed in claim 31 in which the scanner is configured to provide visible light and UV images, in which the visible light image is used to identify the marked pixels and the UV image is used for calculation of tile image densities.

33. An apparatus for testing a document substantially as herein described with reference to, and as illustrated in, the accompanying drawings.

34. A computer-implemented method of testing a digital image of a cheque substantially as herein described with reference to, and as illustrated in, the accompanying drawings.