GB2414297A - Assessing the quality of an image of a cheque - Google Patents

Abstract

A scanned digital image of a cheque is analysed to determine the quality of the image, as a means to evaluate the quality of the scanning or printing process or to verify that the cheque is genuine. The scanned image may be analysed for a descriptor, such as the run lengths of characters in the text, i.e. the lengths of the horizontal or vertical lines making up the letters. Regions of the cheque may be analysed and histograms plotted of run lengths within the regions. Comparisons of the run length histograms against a template can be used to indicate the quality of the image. The template may be built up from an analysis of many cheques, which allows the analysis to account for acceptable variations in quality. Analysis based on descriptors such as run length allow the image quality to be assessed on smaller amounts of data than in prior art methods that use the rasterised version of the scanned image.

Description

A METHOD FOR THE ASSESSMENT OF THE QUALITY AND USABILITY

OF DIGITAL CHEQUE IMAGES

FIELD OF THE INVENTION

This invention relates to a method for the assessment of the quality and usability of digital cheque images. A chcquc is any valuable document that is scanned, where the scanned image of the document needs to be assessed for quality purposes. An example of a cheque is that defined as a check in the US legislation The Check Clearing for the l0 21't Century Act.

DESCRIPTION OF THE PRIOR ART

l5 The advent of 'I'he Check Clearing for the 215 Century Act ('Check 21') in the USA, which is expected to encourage banks to use images of cheques rather than the actual paper documents, has been the stimulus for much research into the nature of Images and the level of authenticity which can be ascribed to them. If an image is to be a satisfactory replacement for a paper document in terms of its processing capabilities, the semantic information that is retrievable from the image must match that which Is retrievable from the original.

complication is the fact that cheques or images of cheques are typically processed at high speeds for the purposes of clearing. 'Ihis implies the use of high speed scanners, converting paper documents into electronic equivalents. Whether this process occurs at a local branch or at a central clearing house, the essential feature Is that no human intervention should be required because of the adverse Impact on the rate of processing.

This invention is concerned with a process whereby the images produced by scanners are assessed for quality using algorithms that function at very high speeds on standard desktop computers.

An existing patent, US 6,363,162 describes a method of accomplishing a similar objective in assessing quality, although it is primarily focussed on the inspection of signatures.

There is less emphasis on computational simplicity and the algorithms described are more time consuming than the method described in this invention.

Many patents have been issued concerning image quality, but these have either been generic in a very broad sense covering such issues as overall contrast, or very specific In measuring how well images conform to a totally prescribed format.

SUMMARY OF THE INVENTION

The invention is a method for assessing the quality of a digital Image of a cheque; comprising the steps of: (a) scanning a particular cheque to generate a scanned, digital image; (b) generating a representation of the scanned image using a descriptor that identifies a visible kind of feature that is potentially present in a localised region of the scanned Image, the representation occupying less data space than a rasterised version of the scanned image; (c) automatically assessing the quality of that particular, scanned image by using the descriptor for the localised region in the scanned image.

As noted above, many patents have been issued concerning image quality, but these have either been generic in a very broad sense covering such issues as overall contrast, or very specific in measuring how well images conform to a totally prescribed format. The present invention bridges the gap between these two poles, describing methods of providing compact and efficient localised descriptors of cheque data to build Actual images of cheques. A virtual image is an abstract representation that describes the appearance of a cheque in mathematical terms; (it does not necessarily correspond to an actual, real image of a cheque).

In one implementation, a descriptor is fully defined by one or more parameters.

Parameters for a specific descriptor can identify different types of visible kinds of features present In the image and distinguish between these kinds of features. Further, the parameters, or the range of values they may take, for a specific descriptor can describe the range in appearance that a feature can potentially take in a image due to variations or defects in the printing or scanning process.

In the preferred implementation, a descriptor is the result obtained usmg a run length analysis of the cheque image. In a run length analysis, a black and white image is uniquely defined by listing the black run lengths and white run lengths traversing the image in a rasterised fashion (row by row.) A letter such as "E" will have several long black run lengths at the top, bottom and middle but rather more short run lengths corresponding to the thickness of the vertical line. Run lengths can also be calculated vertically.

Run length calculations from rasterised images are computationally very simple and hence can be calculated at great speed, an essential requirement for this invention. For example, the scanning can be high speed, automated scanning of bulk quantities of cheques.

One or more parameters of the descriptor constitute a run length profile; the run length profile can then be the histogram of run length frequencies in a given localised region.

Translation of peaks in the histogram corresponds simply to varying the print heaviness or darkness. Different run length profiles, each associated with the descriptor obtained using run length analysis, may be sufficient to distinguish between areas of upper and lower case text. Different run length profiles can also be sufficient to distinguish between streaks and properly formed text.

The quality of the scanned cheque image can be assessed by comparison of descriptors measured from multiple localised regions in the scanned cheque image with a set of standardised descriptors; the standardised descriptors can be derived from a template or calculated by analysis of a representative set of cheques.

The locations of the localised descriptors can be used to further verify the authenticity of the cheques by assessing their conformity to a known template or design of cheque.

The run length profiles of different localised regions are used to assign each region to one of a range of categories corresponding to the likely nature of text or image of which the region is a part.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention will be described with reference to the accompanying Figure 1, in which the first 3 bar charts show the run length profile of the text of the payee on a chcquc as the heaviness of the printing decreases and the two modal values move leftwards. The 4th bar chart shows the more diffuse run length profile of a hand written signature.

DETAILED DESCRIPTION

Overview The present invention descries methods of providing compact and efficient localised descriptors of cheque data to build virtual representations or images of cheques. The descriptors are able, by means of a small number of parameters, to describe both the range of the kinds of features that may occur on the documents in question and also the range in appearance that those features can take, brought about by variation In the printing and scanning processes. The virtual images have smaller amounts of data than the raw rasterised formats - i.e. the mathematical abstract representation is more compact than the rasterised format, leading to computational efficiencies.

The invention utilises the descriptors to assess the quality of images by recognising, for instance, where a descriptor corresponds to unacceptably heavy text, or to features, such as streaks, which may arise from badly maintained scanners.

The descriptors are also used to compare images with known templates In order to detect fraud. In this case, the descriptors are supplemented by information about the location of features. Rather than using templates, descriptors may also be derived from a set of images, so that any incoming image may be assessed to determine whether it lies within the tolerances of the set in question.

Environment An outline of an embodiment of the invention is given here, but it should be understood that there arc obvious related environments where the invention is equally applicable or at least applicable with a minimal amount of adjustment.

In a typical scenario, particularly after the coming into force of Check 21, a paying bank receives digital images of many thousands of cheques from banks of first deposit, reconverting banks and cheque cashing outlets. There is a requirement that such cheques be processed within a narrow time frame and failure to identify irregularities of any sort means that the liability for the transaction's correctness passes to the receiving bank. A first requirement, therefore, is that the images shall contain the required information in a form that can be handled by automatic means, the cost of a human operator being prohibitive. s

A second requirement arises from the need to assess the probability that cheques are counterfeits by investigating whether or not the cheque image could have derived from the known cheque design for that particular account.

This invention comprises a solution to the problem of assessing image quality In such circumstances. Essentially what is presented is an algorithm for processing digital images at high speed to assess whether or not there is clear textual or symbolic information present.

The invention also comprises a solution to the problem of assessing the conformity of cheque images to a known design.

In a typical environment, digital images of cheques are presented in a variety of file formats, TIFF, BMP, JPEG etc., the common feature being that all can be converted into rasterised files representing the cheque. The images may be greyscale, having values for each pixel ranging from O to 255, or black and white, having just two possible states for each pixel. In the case of greyscale images a conversion to black and white takes place before the algorithm is applied, this being achieved by use of a simple threshold decision for each pixel.

In the case of cheques, the font sizes of text are restricted to a fairly narrow range, with the possible exceptions of the text to identify the Bank on whom the cheque is drawn.

The resolution of the scanners (dots per inch) is also restricted by the requirements of the system and this necessarily restricts the variation in the images. The invention Is able to deal with virtually an infinite range of printing and scanning but in any specific instance of its application it is advantageous to be aware of the parameters involved.

Descriptors The fundamental process involved in the mathematical algorithm used in this invention is the production of localised descriptors which are considerably more compact than the original rasterised files obtained by scanning cheques.

The descriptors are fully defined by a small number of parameters and have two particularly important properties. First, the varablllty of the parameters must be such that any features which might appear on a cheque can be distinguished one from another by the parameters to which they correspond. Thus a cheque might have text in upper case or lower case and the parameters for these cases should be sufficiently distinct to allow discrimination between the two cases. Equally, the parameters for a logo or similar feature should differ appreciably from those for text or those for unintended features such as blobs and streaks.

Secondly the descriptors should be able to describe the range in appearance of different features that may occur as a result of the printing and scanning processes. If, for instance, an original electronic image is prmted with more than the average amount of toner and then subsequently scanned on a scanner whose threshold is set to a low value, then text will have a dark appearance. Nonetheless, it should still be recogmsed as corresponding to the same text when imaged in different conditions. A feature of the descriptors must be that the sets of parameters describing these two cases have properties that distinguish them from sets of parameters corresponding to other features.

Likewise, if a cheque is printed or scanned at a different resolution the parameters of the descriptors must still be recognizable as belonging to the same set.

The process of producing a virtual image takes place by inspecting the image to find localised regions or zones where there is significant data, grouping the zones in some way and then deriving descriptors for each zone. From these it should be possible to assess what text is present, whether or not it lies within the acceptable range of darkness or lightness or whether there are features that are purely contingent on some failure of the mechanical processing.

If in addition the cheques are required to fit a known design, the descriptors, can be attached to prescribed locations (e.g. from a known template or design of cheque) and the failure of a chequc to contain descriptors with the correct parameters at the correct locations is taken as prima facie evidence for the existence of a counterfeit.

Run Length Profiles s One possible set of such descriptors is produced by the well known process of describing the image in "run lengths." That is to say a black and white image is uniquely defiecd by listing the black run lengths and white run lengths traversing the image in a rasterised fashion (row by row.) A letter such as "E" will have several long black run lengths at the top, bottom and middle but rather more short run lengths corresponding to the thickness of the vertical line. Run lengths can also be calculated vertically.

The descriptor then is essentially the histogram of the run length frequencies taken in a particular locality.

Run length calculations from rasterised images are computationally very simple and hence can be calculated at great speed, an essential requirement for this invention.

In one implementation, the whole image is subdivided into small, localiscd regions or segments whose width would be about 3 letters and whose height 2 letters, depending on font size. The black and white run lengths within each segment are calculated. The frequencies of the run lengths are used to provide a profile for each segment.

In whatever implementation is selected, the result will be a representation or virtual image consisting of sets of parameters derived from run length analysis.

Different formats of text are found to have different characteristic profiles. All text profiles are predominantly Modal, with one mode corresponding to the width of lines and the other corresponding to the lengths of horizontal features in characters. Text with serifs, for instance, may have a smaller mode arising from the thicknesses of the serifs; larger fonts tend to have higher modes.

Thus if run length profiles are used as descriptors, the transformation required to provide a descriptor for darker text is simply a translation of the peaks of the frequency histogram to the right whilst retaining the original distribution. Figure 1 shows three profiles of the same piece of text from a cheque where the threshold set for the scanner is varied, and as a comparison a profile for a hand written signature is included. It is readily apparent that by simple inspection of the profile the text can be discriminated from the signature and in addition the heaviness of the text can be assessed.

Similarly, if the text is interspersed with artefacts such as small black spots, or is blurred, the run length profile will have recognisably the same shape but will have less contrasting peaks and hollows.

This should be compared with descriptors based purely on frequency analysis. The frequencies give quite a good indication of the presence of text but have a less predictable behaviour if the text becomes darker or lighter.

Another example occurs in patent PC1'/GB2003/001614(HOTDI) where descriptors are provided to determine the precise matching of documents but the descriptors described are deliberately chosen so as to vary according to different environments.

Prior to analysing cheques, analysis of the run length profiles corresponding to a range of fonts and, if required, structured symbols, is deduced by using sets of sample images.

These samples should include cases where the printing is very heavy, right up to the situation where small white areas of characters are filled in, and similarly cases where the printing is very light. Likewise the effect of changing thresholds on scanners should be measured in terms of its effects on run length profiles.

By these means a comprehensive set of profiles of standardised descriptors should be established; when analysing a cheque, the profile selected IS the one that is the most appropriate to the given printing and scanning environment and to the nature of the images In terms of cheque design. The set of parameters within any run length profile will be an indicator both of the type of feature which it purports to describe and of the quality of that feature in terms of blurring, darkness etc. The process of quality assessment or verification of a cheque is implemented by comparing the profiles measured from the cheque being tested, with the previously categorised run length profiles, and assessing the level of correlation. As a result of such a process, those parts of the cheque which contain usable text can be identified, 1nitlally as a set of small segments, where the borders of the segments are not precisely at the borders of the text. By a simple automatic inspection process adjacent segments can be linked together to form larger areas of text.

From this process there will be immediate information as to whether or not there are areas of particularly dark or particularly light print, indicating an image that may not be of acceptable quality. Run length profiles will also reveal the presence of blobs or streaks which may arise from the scanning process.

In one implementation, the run length profiles of the locallsed regions or segments are used to assign each segment to one of a range of categories corresponding to the likely nature of text or image of which the segment is a part. Thus there might be a category for good quality text, a category for areas with many small run lengths arising from the presence of small dots, and a category for many long run lengths corresponding to dark areas. By this means a virtual image or representation of the cheque can be formed whose values are the category values. Any one of these segments may be inaccurately assigned on account of the limited sample size, but the accumulation of values will give a good indication of the nature of the underlying image or text.

Thus, a set of segments corresponding to a dark category might indicate the presence of heavy inking or a scanning artefact, whereas a grouped set of segments classified as text might indicate the position of the textual data and provide a starting pomt for character recognition procedures. A search may similarly be made for other features, using only the virtual image of category values and thus drastically reducing computational requirements.

Quality Assessment At one level of quality assessment, all that is required is verification that there are zones in all parts of the cheques which contain good quality text. This process is simply carried out using the descriptors to verify that the run length profiles do indeed conform to those of the acceptable type and quality of text. At the same time it may be a requirement to identify any unintended features such as streaks, and this also can be done by examining the run length profi1cs of every nonempty zone. This process might be a precursor to closer examination by optical character recognition methods (OCR) of identified text areas.

Cheque Verification By including information about the location of the descriptors it is possible to take quality assessment to a higher level.

In the case of cheques there arc regulations affecting the overall format. Various zones are required to contam various types of information. As an instance, a zone on the right hand side of the cheque is used to record the courtesy amount, that is to say, the amount payable on the cheque encoded as figures rather than words. The run length profile information for figures will be different from that corresponding to text and it IS these figure profiles that will be used in the zone in question.

Typical information to be conveyed by a cheque includes a Bank routing number, the amount, the account number and these all appear on the "MICR" line in a completely defined font at the bottom of the cheque. The name of the Bank on which the cheque IS drawn and the number of the check often appear on preprinted stock. I'he amount (m words and in numbers), and the payee name are printed at time of Issue or handwritten in the case of personal cheques.

It is thus possible to assess whether or not appropriate run length profiles exist In each area of the cheque where the embedded information is expected. A measure of the overall quality of the imaging can be obtained by assessing the presence of good text in areas where it should be present and assessing whether any additional located data is indicative of serious imaging defects or is merely superficial noise.

In another possible implementation, the expected locations of data can be specified with complete precision rather than merely being confined within broad bounds. This Implementation is required where there is an attempt to detect counterfeit cheques because fraudstcrs frequently fail to achieve complete accuracy when copying cheque designs.

This detection can be carried out by providing a template derived from the original cheque design. The template will comprise two types of item. First, there are items which are identical in size and position on every cheque produced for a given account.

These items include the Bank's name and possibly logo, certain words such as 'pay,' date' etc and other decorative elements and are usually printed on the original cheque stock. Descriptors for these items will be very precise. Secondly there ate items which are printed at the time of cheque issue such as the payee and value of the cheque. Now these latter items will have variable size and content but nonetheless will be fairly closely described by descriptors because they will be printed with a known font.

In another implementation, the template described above is derived by analysis of a set of cheques prior to being used for cheque verification. This analysis produces a set of localised descriptors, identifying those zones which are identical on every cheque and those which a variable. The analysis goes further by identifying the variability in the parameters describing the descriptors. By assessing the deviations it is possible to discover whether or not any selected cheque lies within the acceptable bounds of accuracy. This method is particularly powerful as the template can be updated as more cheques are assessed.

Claims (23)

1. A method for assessing the quality of a digital image of a cheque; comprising the steps of: (a) Scanulng a particular cheque to generate a scanned, digital image; (b) generating a representation of the scanned image using a descriptor that identifies a visible kind of feature that is potentially present in a localised region of the scanned image, the representation occupying less data space than a rasterised version of the scanned image; (c) automatically assessing the quality of that particular, scanned image by using the descriptor for the localtsed region in the scanned image.

2. The method of Claim 1 in which a descriptor is fully defined by one or more parameters.

3. The method of Claim 1 in which parameters for a specific descriptor can Identify different types of visible kinds of features present in the image and distinguish between these kinds of features.

4. The method of Claim 3 in which the parameters, or the range of values they may take, for a specific descriptor can describe the range in appearance that a feature can potentially take m a image due to variations or defects in the printing or scanning process.

5. The method of Claim 4 in which a descriptor is the result obtained using a run length analysis of the cheque image.

6. The method of Claim 5 in which one or more parameters of the descriptor

7. The method of Claim 6 In which the run length profiles of different localised regions are used to assign each region to one of a range of categories corresponding to the likely nature of text or image of which the region is a part.

8. The method of Claim 6 in which the run length profile is the histogram of run length frequencies in a given localised region.

9. The method of Claim 8 in which translation of peaks in the histogram corresponds to varying the print heaviness or darkness.

10. The method of Claim 5 where different run length profiles, each associated with the descriptor obtained usmg run length analysis, are sufficient to distinguish between areas of upper and lower case text.

11. The method of Claim 5 where different run length profiles are sufficient to distinguish between streaks and properly formed text.

12. 'I'he method of Claim 1 where the quality of the scanned cheque image is assessed by comparison of descriptors measured from multiple localised regions in the scanned cheque image with a set of standardiscd descriptors.

13. The method of Claim 12 where the descriptors are obtained using a run length analysis of the cheque image and the quality assessment involves verifying that the resultant run length profiles conform to those of the acceptable type and quality of text.

14. The method of Claim 12 in which the standardised descriptors are derived from a template.

15. 'lihe method of Claim 12 where the standardised descriptors are calculated by analysis of a representative set of cheques.

16. The method of Claim 1 where the descriptors are obtained using a run length analysis of the cheque image and the authenticity of the cheque is verified by assessing that appropriate run length profiles exist in each area of the cheque image where embedded information is expected. '

17. The method of Claim 1 where the locations of the localised descriptors are used to further verify the authenticity of the cheque by assessing their conformity to a known template or design of cheque.

18. The method of Claim 17 where the template is derived by analysis of a set of cheques prior to bemg used for cheque verification.

19. The method of Claim 18 in which the analysis produces a set of localised descriptors, identifying those zones which are identical on every cheque in the set and those which a variable.

20. The method of Claim 19 in which the analysis identifies the variability in the parameters describing the descriptors.

21. The method of Claim 20 in which, by assessing the variability in the parameters, it IS possible to determine whether or not any selected cheque that is scanned lies within the acceptable bounds of accuracy.

22. The method of Claim 21 in which the template is updated as more cheques are assessed.

23. The method of Claim 1 in which the scanning is high speed, automated scanning of bulk quantities of cheques.