HK1141355B - Authentication of security documents, in particular of banknotes - Google Patents

Description

Authentication of security documents, in particular banknotes

Technical Field

The present invention relates generally to the authentication of security documents, particularly banknotes. More precisely, the invention relates to a method for checking the authenticity of security documents, in particular banknotes, wherein authentic security documents comprise security features printed, coated or otherwise provided on the security documents, which security features comprise characteristic visual features inherent to the process used for manufacturing the security documents. The invention also relates to a digital signal processing unit suitable for carrying out part of an authentication method, a device for carrying out said authentication method, a method for producing a security document with the aim of optimizing the authentication of the security document according to said authentication method, and a method for detecting security features printed, applied or otherwise provided on a security document, in particular a banknote.

Background

The counterfeiting of security documents, particularly banknotes, is and continues to be a major concern for industry and economy throughout the world. Most counterfeit banknotes are manufactured using common imaging and printing equipment readily available to any user in the consumer market. The advent of scanners and color copiers, as well as high resolution color printers that utilize widely spread printing processes such as inkjet printing, thermal printing, and laser printing, has made it increasingly easy to manufacture a significant number of counterfeit security documents. Most banknote counterfeits are produced by the imaging and printing equipment mentioned above and may be referred to as "colour copies".

There are also counterfeits "offset counterfeits" of offset prints printed using commercial offset printing presses. These counterfeits are typically printed in screen offset (i.e., with a multi-color screen or wire grid combination as a feature of commercial offset printing) and/or line offset (i.e., without any screen or wire grid combination).

Most genuine banknotes combine the features of high quality printing by intaglio printing, line offset printing with high precision front to back registration, and letterpress printing. In particular, gravure and line offset printing allow the production of high resolution patterns with high print definition. Letterpress printing is commonly used to print variable information, such as serial numbers. Additional printing or processing techniques are also utilized to print or apply other features on the banknote, such as screen printing, foil stamping, laser marking or perforation, and the like.

A skilled person having some process knowledge involved in the context of the manufacture of banknotes and similar security documents therefore has no great difficulty in distinguishing most counterfeit documents from genuine documents. Close viewing of counterfeit documents using simple means such as magnifying glasses often makes it possible to immediately identify characteristic features inherent in genuine security documents, such as the intaglio printed security patterns already mentioned which are present on most banknotes. This however requires some skill and knowledge about security printing which is not necessarily ubiquitous in the public. In fact, most people are relatively vulnerable to counterfeiting, provided that the general appearance of the counterfeit or copy is substantially similar to the appearance of a genuine document. This not only represents a problem in the context of banknote counterfeiting, but also with respect to the counterfeiting of other types of valuable documents, such as checks, stamp taxes, identity documents, travel documents and the like.

Machine-based authentication of security documents, i.e. automatic identification in document processing systems such as vending machines, Automatic Teller Machines (ATMs), banknote acceptors and similar financial transaction machines, is also affected by counterfeiting. In fact, it is not uncommon to find fairly advanced counterfeits of security documents that also replicate machine-readable security features, such as infrared, fluorescent and/or electromagnetic markings, present on genuine documents. In fact, most machine-based authentication systems essentially focus on these machine-readable features with little or no actual visual inspection of the visual security features printed, coated or otherwise disposed on the security document.

In other words, the characteristic visual features inherent to the process for manufacturing security documents (in particular intaglio patterns, line offset patterns, relief patterns and/or light diffractive structures) are rarely used in the context of machine-based authentication.

One exception is the so-called ISARD technology, which was invented and developed by the TNO applied physics research institute on behalf of the netherlands national bank in the late 60 s. ISARD stands for Intaglio Scanning and recognition Device and is based on a measure of the characteristic relief profile of a gravure printed feature. A discussion of this identification principle can be found in, for example, the following papers:

[Ren96]Rudolf L.van Renesse，Optical Inspection techniques for Security Instrumentation″，IS&T/SPIE′s Symposium on Electronic Imaging，Optical Security and Counterfeit Deterrence Techniques I，San Jose，California，USA(January 28-February 2，1996)，Proceedings of SPIE vol.2659，pp.159-167；

[ HeiOO ] Hans A.M. De Heij, De Nederlandsche Bank NV, Amsterdam, The Netherlands, "The design method of Dutch banks", IS & T/SPIE's 12th International Symposium on Electronic Imaging, Optical Security and outer destination Deterference technique III, San Jose, California, USA (January27-28, 2000), Proceedings of SPIE vol.3973, pp.2-22; and

[HeiO6]Hans A.M.de Heij，De Nederlandsche Bank NV，Amsterdam，the Netherlands，″Public feedback for better banknote design″，IS&T/SPIE′sInternational Symposium on Electronic Imaging，Optical Security andCounterfeit Deterrence Techniques Vl，San Jose，California，USA(January17-19，2006)，Proceedings of SPIE vol.6075，607501，pp.1-40.

the ISARD identification principle and the devices for implementing it are also disclosed in patent publications GB1379764 (corresponding to NL7017662), NL 7410463, NL 9401796 and NL 9401933.

The problem with the ISARD method is that it is highly dependent on the degree of wear and use of the document, as well as the presence of wrinkles in the banknote substrate, which factors directly affect the actual relief profile on the intaglio print and its detection by the isarad. The ISARD technique is applied, for example, as a pattern of parallel intaglio printed lines on the 50 netherlands shield "Sunflower" banknote (issued in 1982) and on the currently issued euro banknote (see [ HeiO6 ]). Indeed, ISARD was and is now primarily used by the public to perform nail scratch tests (i.e., by drawing a nail across a pattern of parallel intaglio lines).

Further solutions to counterfeit attacks and possibly to implement machine-based authentication may consist in incorporating a special authentication code in the security document itself, for example by using special marker materials such as rare earth elements contained in ink or embedded paper, or by hiding the authentication code in the printed pattern itself using so-called digital watermarking techniques. However, incorporating a special authentication code in a security document implies special handling of the document during the design and/or manufacturing stage, and corresponding specially designed authentication techniques. This correspondingly increases the burden on the designer and/or the printer to adjust the design process and/or the manufacturing process of the security document and also means that special detection techniques must be used for the purpose of the authentication process.

A solution based on the incorporation of a special code in a printed pattern, such as disclosed in european patent application EP 1864825 a1 (which corresponds to international application number WO2006/106677a1 entering the european stage), discloses a printed matter and a method for extracting information from the printed matter, wherein the information is embedded (or coded) into a printed design, in particular a guilloche pattern, such that the information can be detected by subjecting a sample image of the pattern to a fourier transform. The encoding of this information can be done by spatially adjusting the spacing between the parallel/concentric curve image elements. Such spatial adjustment results in the generation of spectral peaks in the fourier transform spectral image of the sample image of the pattern, which spectral peaks represent the information embedded in the printed design and can thus be decoded. More precisely, according to european patent application EP 1864825 a1, the encoded information is extracted by taking into account the spectral peak intensities.

A disadvantage of this method lies in the fact that special codes have to be embedded in the printed pattern in a specific way to allow decoding. This in turn imposes considerable limitations on the designer who must follow special design rules to design the printed pattern. Indeed, the teachings of european patent application EP 1864825 a1 are substantially limited to embedding information in guilloche patterns, as can be easily seen from the figures of observation EP 1864825 a 1.

For example, the method disclosed in european patent application EP 1864825 a1 is applied to encode information on a personalized document, such as an identification card, driver's license, or the like, which information relates to the owner/holder of the personalized document. Information relating to the owner is encoded into the guilloche pattern printed on the personalized document. When the information embedded in the guilloche pattern is user-dependent, this correspondingly makes it more difficult for counterfeiters to produce similar personalized documents. However, any copy of a personalized document made at a similar resolution to the original will exhibit exactly the same information as the original. Such a method is therefore mainly suitable for the purpose of authenticating security documents intended to carry information relevant to the user, which is not the case for banknotes, for example.

Us patent No. 5,884,296 discloses an apparatus for discriminating image properties in a block region contained in a document image, which involves performing fourier transform based on image data in the block region, and determining a spatial spectrum with respect to the image in the block region. The neural network is configured to output a discrimination result as to whether or not an attribute of the image within the block region is a halftone floating point image based on the spatial frequency spectrum output from the fourier transform. This apparatus is intended in particular for use in digital copiers for the purpose of improving the quality of the image. The device of us patent No. 5,884,296 is more particularly intended for use in the context of copying documents containing a mixture of textual, photographic and/or dot images, the image attributes needing to be processed separately to produce good image quality in the copied document. Us patent No. 5,884,296 does not deal in any way with the problem of authenticating security documents, but rather relates to a solution aimed at improving discrimination between different properties of an image.

European patent application No. EP 1484719 a2 discloses a method for forming a template of a reference document, such as a banknote, and using the template to verify other test documents, in particular to verify currency in an automated teller machine. The method includes using multiple reference documents, such as images of real banknotes, and segmenting each image into multiple segments in a similar manner. Each segment is classified using a class classifier to determine a reference classification parameter. These parameters are used to define threshold reference classification parameters. Thus, verification of the test file is performed by comparing the image of the test file with the generated template, rather than by observing the inherent features of the test file.

There is therefore a need for a simpler and more efficient method, in particular one that does not as such utilise the new design and/or manufacturing processes but rather the inherent features of the security features, in particular the characteristic and inherent features of intaglio printed images, that are generally already present on most genuine banknotes.

Summary of The Invention

It is therefore a general object of the present invention to improve the known methods for checking the authenticity of security documents, in particular banknotes.

More precisely, it is another object of the invention to provide a method which exploits the inherent features of security features, in particular of intaglio printed patterns, which have generally been printed, coated or otherwise provided on security documents.

Another object of the present invention is to provide a solution for achieving a reliable and efficient distinction between a genuine (genuine) security document and a copy or counterfeit thereof.

It is a further object of the invention to provide a solution that can be implemented in an automated document processing system, such as a vending machine, ATM or the like, in a simpler manner than currently known solutions.

These objects are achieved thanks to the solution defined in the claims.

According to the present invention, there is provided a method for checking the authenticity of security documents, in particular banknotes, wherein the authentic security document comprises security features printed, coated or otherwise provided on the security document, the security features comprising characteristic visual features inherent to the process for manufacturing the security document, the method comprising the steps of:

-obtaining a sample image of at least one region of interest of the surface of the candidate document to be authenticated, the region of interest encompassing at least a portion of the security feature;

-digitally processing the sample image by performing a decomposition of the sample image into at least one scale subspace containing high resolution details of the sample image and extracting classification features from the scale subspace, and

-deriving a level of authenticity of the candidate document based on the extracted classification features.

Preferably, the digital processing of the sample image comprises: (i) performing a transform on the sample image to obtain at least one set of spectral coefficients representing fine-scale, high-resolution details of the sample image, and (ii) processing the spectral coefficients to extract classification features.

Even more preferably, the transform is a wavelet transform, advantageously a Discrete Wavelet Transform (DWT) selected from the group comprising for example a Haar wavelet transform, a Daubechies wavelet transform and a Pascal wavelet transform. Any other suitable wavelet transform or derivative thereof may be used.

The processing of the spectral coefficients (referred to as "wavelet coefficients" in the context of wavelet transforms) preferably comprises performing a processing of a statistical distribution of the spectral coefficients. The statistical processing may particularly comprise calculating at least one statistical parameter selected from the group comprising an arithmetic mean (first moment in statistics), a variance (second moment in statistics), a skewness (third moment in statistics), an excess (fourth moment in statistics) and an entropy of the statistical distribution of said spectral coefficients.

The decomposition of the sample image is advantageously performed as a result of one or more iterations of a multi-resolution analysis (MRA) of the sample image.

According to the present invention, there is also provided a method for checking the authenticity of a security document, in particular a banknote, wherein the authentic security document comprises security features printed, coated or otherwise provided on the security document, which security features comprise characteristic visual features inherent to the process used to manufacture the security document, said method comprising the step of digitally processing a sample image of at least one region of interest of the surface of a candidate document to be authenticated, the digital processing comprising performing one or more iterations of a multi-resolution analysis of the sample image.

The above method may be provided for digitally processing multiple sample images corresponding to several regions of interest of the same candidate file.

According to a preferred embodiment of the invention, the sample image can be obtained at a relatively low resolution, i.e. below 600dpi, preferably at 300 dpi. In fact, tests have shown that a high scanning resolution for the sample image is not necessary at all. This is particularly advantageous because the low resolution reduces the time necessary to perform the acquisition of the sample image and reduces the amount of data to be processed for a given surface area, which in turn substantially facilitates the practical implementation of the method.

Within the scope of the invention, the security features used for authentication mainly comprise intaglio patterns. However, the security features may include intaglio patterns, line offset patterns, relief patterns, light diffractive structures (i.e., patterns or structures inherent to the process performed by the security printer), and/or combinations thereof.

Maximization of the qualification level is achieved by ensuring that the selected region of interest comprises a high density (high spatial frequency) pattern, preferably a gravure printed straight or curved pattern. These patterns may in particular be patterns of a pictorial representation, such as a portrait, arranged on the candidate document.

Also claimed is a digital signal processing unit for processing image data of a sample image of at least one region of interest of a surface of a candidate document to be authenticated according to the above method, the digital signal processing unit being programmed to perform digital processing of the sample image, the digital signal processing unit being advantageously implementable in an FPGA (field programmable gate array) unit.

A device for checking the authenticity of security documents, in particular banknotes, according to the above method is similarly claimed, which device comprises an optical system for obtaining a sample image and a digital signal processing unit programmed to perform digital processing of the sample image.

Further claimed is a method for manufacturing a security document, in particular a banknote, comprising the step of designing a security feature printed, coated or otherwise provided on said security document, wherein the security feature is designed to optimize the level of authenticity calculated according to the above method by generating a characteristic response in said at least one dimensional subspace.

The use of wavelet transforms and multi-resolution analysis for authenticating security documents, in particular banknotes, is also claimed.

Finally, there is provided a method for detecting security features printed, coated or otherwise provided on security documents, in particular banknotes, which security features comprise characteristic visual features inherent to the process for manufacturing the security documents, said method comprising the step of digitally processing a sample image of at least one region of interest of a surface of a candidate document, wherein the region of interest is selected to encompass at least a portion of said security features, the digital processing comprising performing one or more iterations of a multi-resolution analysis of the sample image to extract classification features characterising said security features. This method is particularly advantageously applied to the detection of intaglio printed images.

Brief description of the drawings

Further characteristics and advantages of the invention will become better apparent from reading the following detailed description of embodiments of the invention, which are illustrated by way of non-limiting example only and are illustrated in the accompanying drawings, wherein:

FIG. 1a is a gray scale scan of an exemplary banknote sample;

FIG. 1b is a grayscale photograph of the upper right hand portion of the banknote specimen of FIG. 1 a;

fig. 2a and 2b are enlarged views of a banknote specimen of fig. 1a, fig. 2b corresponding to the area indicated by white squares in fig. 2 a;

fig. 3a and 3b are enlarged views of a first color reproduction of the banknote specimen of fig. 1a, fig. 3b corresponding to the area indicated by the white squares in fig. 3 a;

fig. 4a and 4b are enlarged views of a second color reproduction of the banknote specimen of fig. 1a, fig. 4b corresponding to the area indicated by the white square in fig. 4 a;

FIG. 5a is a schematic diagram of a one-level (one iteration) discrete wavelet transform;

FIG. 5b is a schematic diagram of a three-level (three iterations) discrete wavelet transform;

FIG. 6 is a schematic diagram illustrating the principle of multi-resolution analysis (MRA);

FIG. 7a shows one iteration of a two-dimensional wavelet transform;

FIG. 7b shows a second iteration of the two-dimensional wavelet transform after the first iteration shown in FIG. 7 a;

FIG. 8 is a schematic diagram of a so-called "non-standard decomposition" method for performing a two-dimensional wavelet transform;

FIG. 9 is a schematic diagram of a so-called "standard decomposition" method for performing a two-dimensional wavelet transform;

FIG. 10a is a graphical representation of the results of one iteration of a two-dimensional wavelet transform applied to image data corresponding to the region of interest shown in FIG. 2 b;

FIG. 10b is a graphical representation of the results of one iteration of a two-dimensional wavelet transform of image data corresponding to the region of interest shown in FIG. 2b, as shown in FIG. 10a, with detailed sub-images normalized for better visual representation;

11a to 11c are 3 illustrations of the combined results of detailed sub-images (as shown in FIG. 10 b) normalized for better visual representation, where FIGS. 11a, 11b and 11c show the results of the processing of the images of FIGS. 2b, 3b and 4b, respectively;

fig. 12 shows 9 histograms showing the statistical distribution of the wavelet coefficients resulting from a one-level wavelet transform of the images of fig. 2b, 3b and 4b, the histograms of the upper, middle and lower lines of the 3 histograms representing the horizontal, vertical and diagonal details resulting from said wavelet transform, respectively;

FIG. 13 is a diagram of two statistical parameters, skewness (also known as third moment in statistics) and excessive kurtosis (also known as fourth moment in statistics), which can be used to characterize the statistical distribution of wavelet coefficients;

14a to 14c are 3 bar graphs showing the variance of the statistical distribution of wavelet coefficients, i.e., a measure of dispersion, resulting from one-level wavelet transforms of the images of FIGS. 2b, 3b and 4b for horizontal, vertical and diagonal details, respectively;

figures 15a and 15b are two enlarged views of a portion of the intaglio printed portrait of Bettina von Arnim appearing on the front face of a 5 mark banknote issued in germany during 1991 to 2001 before being introduced in euro;

figure 16a is a view showing 6 grey scale scans of substantially the same area of 2 original samples (illustrations a and B) and 4 colour copies (illustrations C to F) of a 5 mark banknote;

FIG. 16b shows 6 histograms showing the statistical distribution of wavelet coefficients resulting from a one-level wavelet transform of the image of FIG. 16a, each histogram showing the statistical distribution of combined wavelet coefficients (i.e., a combination of horizontal, vertical and diagonal details);

FIG. 17 is an illustrative overlay of the histograms of the upper left and lower right corners of FIG. 16 b;

fig. 18a is a bar graph showing the variance of the statistical distribution of wavelet coefficients resulting from a one-level wavelet transform of image data corresponding to the same region of interest (as shown in fig. 15b and 16 a) as 11 candidate documents, these 11 candidate documents comprising 5 original samples of 5 mark banknotes (candidate documents 1 to 5) and 6 color copies (candidate documents 6 to 11);

fig. 18b is a bar graph showing the excessive kurtosis, i.e. a measure of "kurtosis", of the statistical distribution of wavelet coefficients resulting from one-level wavelet transform of image data corresponding to the same region of interest as the same 11-mark banknote candidate documents as in fig. 18 a;

FIG. 19 is a schematic representation of an exemplary feature space for classifying candidate documents, where the variance and excessive kurtosis of the statistical distribution of wavelet coefficients are used as (x, y) coordinates to locate candidate documents in the feature space;

FIG. 20 is a schematic representation of an exemplary feature space similar to that of FIG. 19, wherein a plurality of candidate files including an original sample and a color replica are represented in the feature space using variance and excessive kurtosis as (x, y) coordinates;

figure 21 is a schematic view of an apparatus for checking the authenticity of a security document according to the method of the invention; and

FIG. 22 is a generalized flow diagram of a method according to the present invention.

Detailed description of embodiments of the invention

The present invention stems from the observation that security features (particularly intaglio printed features) printed, coated or otherwise provided on security documents using a particular manufacturing process that is only achievable with security printers exhibit highly characteristic visual features (hereinafter referred to as "intrinsic features") that can be identified by qualified persons having knowledge about the particular manufacturing process involved.

The following discussion will focus on analyzing the inherent features produced by gravure printing. It will be appreciated, however, that the same method is also applicable to other intrinsic features of banknotes, in particular offset line printing features, letterpress features and/or light diffractive structures. The results of the tests carried out by the applicant show that the intaglio printing features are well suited for the purpose of the authentication according to the present invention and further give the best results. This is due in particular to the fact that gravure printing enables the printing of very fine, high resolution and sharp patterns. Gravure printing is therefore the preferred process for making the inherent features utilized in the context of the present invention.

Figure 1a is a grey scale scan of an exemplary banknote sample 1 showing a portrait of Jules Verne made by the applicant during 2004. The banknote specimen 1 is manufactured using a combination of printing and processing techniques specific to banknote manufacture, in particular, these printing and processing techniques include line offset printing for printing a multicolour background 10 of the banknote, screen printing for printing optically variable ink patterns including patterns of a planar celestial figure 20 and a sextant 21, foil stamping techniques applying optically variable devices, a strip of material 30 including an optically diffractive structure carrying optical diffractive structures extending vertically along the height of the banknote (the strip of material 30 being schematically delimited in fig. 1a by 2 dashed lines), intaglio printing for printing several intaglio patterns 41 to 49 including a portrait 41 of jules verne, relief printing for printing 2 serial numbers 51, 52, and varnishing for varnishing the banknote with a layer of protective varnish. The banknote sample 1 is also provided with indicia 60 on the right hand edge of the sample, the indicia 60 being applied by laser ablation of portions of the strip 30 and the underlying layer of offset printing ink (not referenced). In the example shown, a portrait 41 (together with the vertical year name 2004 and the visual graphics around the portrait), the symbol "KBA-GIORI" with the flying horse 42, the symbols "KBA-GIORI" 43 and "specialen" 44, and tactile patterns 45 to 49 on the three corners of the banknote and on the left-hand and right-hand sides of the banknote are printed by intaglio printing on top of the row offset printed background 10, the screen printed graphics 20, 21 and the strip of material 30. After the intaglio printing stage, the serial numbers 51, 52 are printed and the varnishing is completed. It should be further understood that banknote samples 1 are manufactured on a single sheet of printing and processing equipment (as provided by the present inventors), each sheet carrying an array of multiple banknote samples (as is common in the art) which are ultimately cut into individual banknotes at the end of the manufacturing process.

Fig. 1b is a grayscale photograph of the upper right corner of the banknote specimen of fig. 1a showing in more detail the intaglio printed logo "KBA-GIORI" with the zebra 42 and the tactile pattern 45 comprising a set of parallel lines partially overlapping the zebra 42 at a 45 degree angle. The characteristic embossing and relief effect of the intaglio print and the clarity of the print can be easily seen in this photograph.

FIG. 2a is a left-hand portion of the portrait 41 of FIG. 1a (with the patterns 20, 21 and 44 in view)2 a) is also partially visible). Fig. 2b is an enlarged view of a square portion (or region of interest r.o.i.) of the portrait 41 shown by the white square in fig. 2 a. Figure 2b shows some characteristic intrinsic features of the intaglio patterns constituting the portrait 41. The region of interest r.o.i for subsequent signal processing does not need to cover a large surface area of the document. More precisely, tests have shown that less than 5cm²Already sufficient surface area for authentication purposes.

Figures 3a, 3b and 4a, 4b are greyscale images similar to figures 2a, 2b of 2 colour copies of the banknote specimen shown in figure 1a, which were made using commercial colour copying equipment. In each of fig. 3a and 4a, the depicted white squares indicate the respective region of interest r.o.i of the portrait displayed in the enlarged view in fig. 3b and 4b, respectively. The first color reproduction shown in figures 3a, 3b was made using an Epson ink jet printer and Epson photographic paper. A second color reproduction shown in fig. 4a, 4b was made using a Canon ink jet printer and plain paper. A high resolution scanner is used to scan the original sample and provide the necessary input to the inkjet printer.

While the overall visual aspect of the 2 color copies appeared similar to the original sample, as shown in fig. 3b and 4b, looking more closely at the structures of the intaglio patterns constituting the portrait that were replicated showed that the structures were not as sharp as in the original sample (see fig. 2b) and that the structures appeared somewhat hazy and smooth due to the inkjet printing process and the nature of the paper used. The image information contained in fig. 3b and 4b is clearly different from the image information of the original sample shown in fig. 2 b. The invention therefore relates to a method of defining how to display and use this difference in order to distinguish between original and authentic samples in figures 2a, 2b and duplicates of figures 3a, 3b and 4a, 4 b. The following discussion will focus on this problem.

As alluded to above, the inherent and characteristic features of intaglio printed patterns are in particular a high definition of the printing, whereas inkjet printed copies exhibit a substantially lower print definition, in particular due to digital processing and printing. This is also the case for color laser printed copies and copies obtained by thermal sublimation processing. This difference can be shown in the following way: the image data (or region of interest) contained in the enlarged view of the candidate document to be identified, such as the views of fig. 2b, 3b and 4b, is decomposed into at least one scale subspace containing the high resolution details of the image, and representative classification data is extracted from the scale subspace, as will be explained in more detail below.

Preferably, the decomposition of the image is achieved by performing digital signal processing techniques based on so-called wavelets (the french term). Wavelets are mathematical functions used to separate a given function or signal into different scale components. Wavelet transform (or wavelet transform) is a representation of a function or signal by a wavelet. Wavelet transforms have advantages over conventional fourier transforms for representing functions and signals that have discontinuities and sharp peaks. According to the invention, the characteristic of the so-called Discrete Wavelet Transform (DWT) is used in particular, as will be discussed below.

It should be appreciated that fourier transforms (as used in the context of the solutions discussed for example in european patent application EP 1864825 a1 and US patent No. US 5,884,296) are not equivalent to wavelet transforms. In fact, the fourier transform involves only converting the processed image into a spectrum representing the relevant spatial frequency content of the image, without any difference in scale.

The wavelet theory will not be discussed in depth in this description, as this theory is also well known in the art and is discussed and described in detail in some textbooks on this subject. The interested reader may refer to the following books and papers on wavelet theory, for example:

[Mal89]Stephane G.Mallat，″A Theory for Multiresolution SignalDecomposition：The Wavelet Representation″，IEEE Transactions on PatternAnalysis and Machine Intelligence，Vol.11，No.7(July 7，1989)，pp.674-693；

[Dau92]lngrid Daubechies，″Ten Lectures on Wavelets″，CBMS-NSFRegional Conference Series in Applied Mathematics 61，SIAM(Society forIndustrial and Applied Mathematics)，2^nd edition，1992，ISBN 0-89871-274-2；

[Bur98]Sidney C.Burrus，Ramesh A.Gopinath and Haitao Guo，″Introduction to Wavelets and Wavelet Transforms：A Primer″，Prentice-Hall，Inc.，1998，ISBN 0-13-489600-9；

[Hub98]Barbara Burke Hubbard，″The World According to Wavelets：The Story of a Mathematical Technique in the Making″，A K Peters，Ltd.，2^ndedition，1998，ISBN 1-56881-072-5；

[Mal99]MALLAT，Stephane，″A wavelet tour of signal processing″， Academic Press，2^ndedition, 1999, ISBN 0-12-466606-X; and

[WalO4]WALNUT，David F.″An Introduction to Wavelet Analysis″，Birkhauser Boston，2^nd edition，2004，ISBN 0-8176-3962-4。

it is understood that wavelets can be generated by a wavelet function (or "mother wavelet") ψ and a scale function (or "parent wavelet")A convenient representation is sufficient. In practice, the wavelet function ψ can be expressed as a band-pass/high-pass filter which filters the upper half of the signal scale/spectrum while the scale function isWhich may be represented as a low pass filter, filters the remaining lower half of the signal scale/spectrum. This principle is schematically illustrated in fig. 5a as a one-stage digital filter bank comprising a filter bank havingA low-pass filter with function h (n) and a high-pass filter with function g (n) that divide the signal scale/spectrum into two parts of equal spectral range. When passing the discretely sampled signal x (n) through the filter bank in fig. 5a, we can consider a one-level wavelet transform of the sampled signal x (n). Output y of a low-pass filter which is essentially the result of convolution of the signal x (n) with a function h (n) (-)_LOW(n) the output y of the high-pass filter comprising the scaling function transform coefficients, or simply "scaling coefficients" (also called approximation coefficients), and similarly the result of the convolution of the signal x (n) with the function g (n) (-)_HIGH(n) include wavelet function transform coefficients, or simply "wavelet coefficients" (also referred to as detail coefficients).

Because each filter filters half of the spectral components of signal x (n), half of the filtered samples may be discarded according to nyquist's law. In fig. 2, the output of the low-pass and high-pass filters is thus downsampled to half (so the downsampling operator "↓ 2" follows each filter in fig. 5 a), which means that every 2 samples are discarded one.

Following this approach, the signal can be decomposed into a plurality of wavelet coefficients corresponding to different scales (or resolutions) by iteratively repeating the process, i.e., by passing the approximation coefficients output by the low-pass filter to a subsequent similar filtering stage. This method is called multiresolution analysis or MRA (see [ Mal89]), the case of three-level multiresolution analysis being schematically shown in fig. 5 b. As can be appreciated in fig. 5b, the filter bank is in fact a three-stage filter bank, wherein the low-pass filtered output of a preceding filtering stage is filtered again by a subsequent filtering stage.

In fig. 5b, the signal x (n) is effectively decomposed into 4 signal components corresponding to 3 different scales, i.e., (i) detail coefficients at a first scale (first order coefficients) comprising half the number of samples compared to the signal x (n), (ii) detail coefficients at a second scale different from the first scale (second order coefficients) comprising 1/4 number of samples compared to the signal x (n), and (iii) approximation coefficients and (iv) detail coefficients at a third scale (third order coefficients), each comprising 1/8 number of samples compared to the signal x (n).

In fact, a discrete sampled signal can eventually be completely decomposed into a set of detail coefficients (wavelet coefficients) at different scales, as long as the sampled signal contains 2^NA number of samples, where N is the number of iterations or stages required to completely decompose the signal into wavelet coefficients.

In summary, multiresolution analysis (MRA) or multiscale analysis refers to a wavelet transform based signal processing technique whereby, as schematically shown in the diagram of fig. 6, a signal is decomposed in a plurality of nested subspaces of different scales ranging from fine details (high resolution components) to coarse details (low resolution components) of the signal.

According to the invention, the intrinsic features of genuine security features, in particular the intrinsic features of intaglio patterns, will be identified by looking specifically at the fine high resolution (fine scale) details of the image of the candidate document to be authenticated, rather than looking at the coarser low resolution details of the image of the candidate document.

Up to now, the wavelet theory has been discussed only in the context of processing one-dimensional signals. However, the image is treated as a two-dimensional signal, which accordingly requires two-dimensional processing. The concept of a two-dimensional wavelet transform will therefore be briefly discussed before turning to the actual description of the preferred embodiments of the present invention.

The wavelet theory discussed above can be easily extended to the decomposition of two-dimensional signals, as discussed for example in [ Mal89 ]. Two-dimensional wavelet transforms basically involve the processing of two-dimensional signals in the row and column directions, where the rows and columns of signals are processed separately using the one-dimensional wavelet algorithm discussed above. This will be explained with reference to fig. 7a, 7b, 8 and 9.

In fig. 7a, an original image (i.e. an image of a selected region of interest of a sample image corresponding to a candidate document to be authenticated-such as the images of fig. 2b, 3b and 4b) is schematically shown, which original image is appliedIs denoted by c⁰. The original image c⁰Composed of a matrix of n x n pixels, where n may be 2^NN is an integer corresponding to the number N of wavelet iterations desired to be performed. In practice, the size of the image should be large enough to encompass a relatively large number of features. For the sake of illustration, the original image c⁰May for example consist of a matrix of 256x256 pixels. But other image sizes are fully possible. At a sampling resolution of 300dpi, it will be appreciated that such image sizes correspond to approximately 2x2cm on the candidate document to be authenticated²The surface area of (a).

As a result of one iteration of the wavelet transform, the original image c is shown in FIG. 7a⁰Is decomposed into 4 sub-images c¹、d¹ ₁、d¹ ₂And d¹ ₃Each sub-image having a size of (n/2) x (n/2) pixels. Sub-image c¹Including along the original image c⁰Of rows and columns of the original image c⁰An approximation of. On the other hand, the sub-image d¹ ₁、d¹ ₂And d¹ ₃Including from along the original image c⁰And high-pass filtering of the rows and columns of the original image c⁰Details of (a). More precisely:

-d¹ ₁is along the original image c⁰And low-pass filtered along the columns and comprises the original image c⁰Horizontal details of (d);

-d¹ ₂is along the original image c⁰And the results of the row low-pass filtering and the along-column high-pass filtering of (c) and including the original image c⁰Vertical details of (d); and

-d¹ ₃is along the original image c⁰And comprises the result of the row and column high-pass filtering of c, and comprises the original image c⁰Diagonal details of (d).

The process may be performed by similarly dividing sub-image c¹Is decomposed into 4 appendixesSub-image c of²、d² ₁、d² ₂And d² ₃Repeated during subsequent iterations, each sub-image having a size of (n/4) x (n/4) pixels, as schematically shown in fig. 7 b. In FIG. 7b, sub-image d¹ ₁、d¹ ₂And d¹ ₃Representing an image c at a first resolution (or scale)⁰And sub-image d² ₁、d² ₂And d² ₃Representing an image c at a second resolution (or scale)⁰The second resolution is half of said first resolution.

After N iterations, original image c⁰Will thus be decomposed into 3N +1 sub-images d^m ₁、d^m ₂、d^m ₃And c^NWherein m is 1, 2. As already suggested above, sub-image d^m ₁Each will contain the horizontal detail of the original image at a different scale (or resolution), while sub-image d^m ₂And d^m ₃Each containing vertical or diagonal details of the original image at different scales, respectively.

The two-dimensional wavelet transform is preferably performed according to a so-called "non-standard decomposition" method, which is schematically illustrated in fig. 8. According to this decomposition method, one-dimensional wavelet transform is alternately performed on rows and columns of an image. In fig. 8, reference symbols a, D, a, D denote:

a: approximate (i.e., low-pass filtered) coefficients of the lines of the image;

d: the detail (i.e., high-pass filtered) coefficients of the lines of the image;

a: approximate (i.e., low-pass filtered) coefficients of the columns of the image; and

d: the detail (i.e., high pass filtered) coefficients of the columns of the image.

As shown in the upper part of fig. 8The rows of the original image are processed first and then the columns, for example to produce the result shown in FIG. 7a (where Aa, Da, Ad and Dd correspond to sub-image c, respectively¹、d¹ ₁、d¹ ₂And d¹ ₃). As shown in the lower part of fig. 8, sub-image Aa (which corresponds to sub-image c)¹) Starting with a row and then a column are similarly processed, resulting in the same decomposition as shown in FIG. 7b (where Aaaa, AaDa, AaAd and AaDd correspond to sub-image c, respectively²、d² ₁、d² ₂And d² ₃)。

An alternative to the "non-standard decomposition" approach discussed above is the so-called "standard decomposition" approach, which is implemented by performing all required iterations along a row, and then only required iterations along a column. This approach is schematically illustrated in fig. 9.

The advantage of the "standard decomposition" method is the fact that each row and column of the image only needs to be loaded from memory once in order to convert the entire image. This approach requires a correspondingly minimum number of memory accesses, which is advantageous in the context of an FPGA (field programmable gate array) implementation.

Although the "non-standard decomposition" method requires more memory accesses than other methods, it has the advantage of requiring less computation time, since during each iteration only one quarter of the data resulting from the previous iteration has to be processed. Furthermore, as is readily understood from a comparison of fig. 8 and 9, the horizontal and vertical details are extracted separately by a "non-standard decomposition" method.

Different types of Discrete Wavelet Transforms (DWT) are suitable in the context of the present invention. Successful testing was accomplished, in particular, by using the so-called Haar-, Daubechies-and Pascal wavelet transforms, which are also known in the art.

The Haar wavelet transform is in fact the first known wavelet transform. This wavelet transform (although not named identically at the time) was discovered by Alfred Haar, Hungarian, in 1909. This wavelet transform is also considered a special case of the so-called Daubechies wavelet transform. The corresponding high-pass and low-pass filters of the Haar wavelet transform each consist of 2 coefficients, namely:

for the low-pass filter:

and

and a high-pass filter:

and

the Daubechies wavelet transform (see Dau 92) is named by belgium physicist and the collectist Daubechies. Daubechies wavelets are a family of orthogonal wavelets and are characterized by a maximum number of so-called vanishing moments (or branches).

In this family of Daubechies wavelet transforms, for example, a so-called Daubechies 4-branch wavelet (or db4 transform), where the filter coefficients consist of 4 coefficients, namely:

for the low-pass filter:

and

and a high-pass filter:

and

the advantage of the Daubechies-db4 transform over the Haar wavelet transform is in particular the increased filtering efficiency of the Daubechies transform, i.e. the cut-off frequencies of the low-pass and high-pass filters are more clearly defined.

The Pascal wavelet transform is based on binomial coefficients of a Pascal triangle (named by french philosopher and the mathematician blaisepasca). Although the Pascal wavelet transform has a less clearly defined cut-off frequency than the Haar-and Daubechies wavelet transforms, the transform may better approximate a continuous signal than the Haar wavelet transform, requiring less computation time than the Daubechies wavelet transform.

For purposes of illustration, the following Pascal wavelet transform may be used, where the low-pass and high-pass filters are each defined by the following 3 filter coefficients:

for the low-pass filter:

and

and a high-pass filter:

and

in contrast to Haar-and Daubechies wavelet transforms, the Pascal wavelet transform is a non-orthogonal wavelet.

Although Haar-, Daubechies-and Pascal wavelet transforms have been mentioned above as discrete wavelet transforms that may be used in the context of the present invention, these should only be considered as preferred examples. Other discrete wavelet transforms are further known in the art (see, e.g., [ Mal99 ]).

In accordance with the present invention, it should again be appreciated that one is primarily interested in the fine, high resolution details of the selected region of interest of the sample image of the candidate file. In other words, according to the present invention, the signal (i.e., the image data of the region of interest) need not be completely decomposed into wavelet components. Thus, as will be appreciated from the following, in order to extract relevant features that can create representative classification data for a candidate document to be identified, it is sufficient to perform one or more iterations of the wavelet transform on the image data. This means that the most relevant scales of the image to be considered are those corresponding to fine, high resolution details which were first obtained in the course of the multi-resolution analysis.

Tests performed by the applicant have shown that one iteration of the wavelet transform (i.e. a first level resolution analysis as schematically shown in fig. 5 a) is in most cases sufficient to extract the necessary features capable of classifying (and therefore distinguishing) the candidate document being authenticated into the categories of genuine or possibly genuine documents or copied/counterfeited documents. In other words, the sample image may simply be decomposed into at least one fine-scale subspace containing the high-resolution details of the sample image.

However, it is entirely possible within the scope of the invention to perform more than one iteration of the wavelet transform, i.e. to extract sets of detail coefficients (or wavelet coefficients) corresponding to more than one high resolution scale of the image data. For computational and processing efficiency, it is preferable to keep the number of iterations as low as possible. Furthermore, as already explained above, according to the invention, it is not necessary to decompose the signal completely into wavelet components, since the resulting wavelet components that will be obtained correspond to the low resolution, coarse content of the image that is expected to be relatively similar between a genuine document and its counterfeits. In fact, this explains in part why non-technical personnel without special knowledge about security printing may be so easily fooled by the general visual appearance and look of counterfeit documents.

The following discussion will therefore focus on the case of one-level wavelet transform involving only one iteration of a two-dimensional wavelet transform, as schematically illustrated in fig. 7a, i.e. the region of interest will be decomposed into 4 sub-images c¹、d¹ ₁、d¹ ₂And d¹ ₃。

Figure 10a shows the result of a first iteration of a two-dimensional wavelet transform applied to an image as shown in figure 2 of an original banknote sample. In this example, the original image has a size of 252x252 pixels, and the image is processed using the above-mentioned Haar wavelet transform.

Approximation image c from low pass filtering¹Is shown in the upper left corner of figure 10 a. Due to the fact that the wavelet coefficients have small values and also contain negative coefficients (so that the wavelet coefficients appear as substantially "black" pixels when directly visualized), the detail image d resulting from the high-pass filtering¹ ₁、d¹ ₂And d¹ ₃Is displayed as a substantially dark area.

In order to better observe the image d¹ ₁、d¹ ₂And d¹ ₃The images may be normalized such that the coefficients are included in a range of values 0 to 255 (i.e., an 8-bit value range of a grayscale image). Such a view is shown in FIG. 10b, where [ d ]¹ ₁]_N、[d¹ ₂]_NAnd [ d¹ ₃]_NRespectively show detail images d¹ ₁、d¹ ₂And d¹ ₃Normalized form of (a). From observing fig. 10b, it can be seen that the wavelet transform is sufficient to detect sharp transitions of the intaglio pattern.

FIG. 11a shows 3 detail images d from FIGS. 10a, 10b¹ ₁、d¹ ₂And d¹ ₃Normalized image obtained in combination [ d ]¹ _G]_N. FIGS. 11b and 11c show corresponding normalized images [ d ] obtained as a result of wavelet variations of the images of the first and second color copies of FIGS. 3b and 4b, respectively¹ _G]_N。

It can be seen that there is a considerable visual difference between the image of figure 11a and the images of figures 11b and 11 c. In particular, it can be seen that the edges of the pattern appear more clearly in fig. 11a than in fig. 11b and 11 c.

Now that the images of the various candidate documents have been processed, it will be explained how representative features can be extracted from these processed images in order to classify and distinguish these documents.

FIG. 12 is an illustration of 9 histograms showing wavelet coefficients for horizontal, vertical and diagonal details (i.e., detail image d) for each of the images of FIGS. 2b, 3b and 4b¹ ₁、d¹ ₂And d¹ ₃Wavelet coefficients). More precisely, the left, middle and right columns of fig. 12 show the corresponding histograms obtained for the images of fig. 2b, 3b and 4b, respectively, while the upper, middle and lower rows of fig. 12 show the corresponding histograms for horizontal, vertical and diagonal details, respectively.

As can be seen from fig. 12, the histogram obtained from the image of the original sample (left column in fig. 12) is wider than the histograms obtained from the image of the color replica (middle and right columns in fig. 12). In other words, the variance σ²That is, a measure of the dispersion of the wavelet coefficients may be conveniently used to classify the statistical distribution of the wavelet coefficients. Variance σ in statistics²Also known as "second moments". Alternatively, a so-called standard deviation σ, which is the variance σ, may be used²The square root of (a).

Variance σ removal²And the standard deviation σ, additional statistical parameters can also be used to characterize the statistical distribution of the wavelet coefficients, i.e.:

the arithmetic mean of the wavelet coefficients-also referred to in statistics as "first moment";

skewness of the statistical distribution of the wavelet coefficients-also statistically referred to as "third moment" -which is a measure of the asymmetry of the statistical distribution;

-excessive, or excessive kurtosis, (or simply "kurtosis") -also known in statistics as "fourth-order moments" -which are measures of "kurtosis" of the statistical distribution; and/or

Statistical entropy, which is a measure of the variation of the statistical distribution.

For feature extraction purposes, the moments listed above (including variances) should be normalized to achieve proper comparison and classification of various candidate documents.

Fig. 13 shows the concept of skewness and excess. "Positive skewness" (as shown) is understood to characterize a statistical distribution, where the right tail of the distribution is longer and where the "body" of the distribution is centered on the left. The opposite is "negative skewness". On the other hand, "positive/high transition" or "negative/low transition" (as shown) is understood to characterize a statistical distribution comprising a sharper peak and a flatter tail, and a rounder peak and a wider "shoulder", respectively.

In the following, overruns (hereinafter denoted by reference character C) will be utilized in particular together with the variance σ²Together as features for further classification.

FIGS. 14a to 14c are 3 bar graphs showing the variance σ of the statistical distribution of wavelet coefficients shown by the graph of FIG. 12². Reference numerals 1, 2, 3 in fig. 14a to 14c refer to the 3 candidate files processed, i.e., the original sample (fig. 2a and 2b), the first color reproduction (fig. 3a and 3b), and the second color reproduction (fig. 4a and 4b), respectively. In FIG. 14a, the variance σ²Is shown for horizontal detail, while FIGS. 14b and 14c show the variance σ, respectively²For vertical and diagonal details.

As expected, the variance σ in the case of deriving the distribution of wavelet coefficients from the image of the original sample²Substantially higher than the variance calculated in the statistical distribution of the wavelet coefficients obtained from the image of the color reproduction.

Tests were performed on various original (i.e. authentic) banknote samples and colour (i.e. counterfeit) copies thereof. These tests show that the method according to the invention is very reliable, especially when the image data of the region of interest being processed comprises relatively high density intaglio printed features, such as in the case of portrait sections or any other similar density pictorial representation that can be found on most banknotes (such as intaglio printed patterns representing buildings on euro banknotes). These tests also show that areas containing a lower amount of intaglio features still lead to good results.

Fig. 15a and 15b are two enlarged views of a portion of the intaglio printed portrait of Bettina von Arnim appearing on the front side of a 5 mark banknote issued in germany during 1991 to 2001 before being introduced in euro. Fig. 15b shows in particular an example of possible regions of interest for the purpose of identification according to the method described above.

Several candidate documents were tested, including original banknotes with different degrees of wear and color copies of banknotes manufactured using inkjet, dye sublimation and color laser copying and printing equipment. For illustrative purposes, FIG. 16a shows six similar images of the same area of interest taken from an original specimen in very good condition (illustration A), an original specimen with relatively high wear (illustration B), a color reproduction made by inkjet printing on photo quality paper at a resolution of 5600dpi (illustration C), a color reproduction made by inkjet printing on plain paper at a resolution of 5600dpi (illustration D), a color reproduction made by thermal sublimation on photo quality paper at a resolution of 300dpi (illustration E), and a color reproduction made by laser printing on plain paper at a resolution of 1200dpi (illustration F).

Fig. 16b shows the corresponding histogram of the statistical distribution of the wavelet coefficients (in fig. 16b the histogram is obtained from the combination of 3 detail images resulting from the low-pass filtering of the image of fig. 16 a). It can be seen that the histograms calculated from the images of the two original samples (histograms a and B in fig. 16B) are highly similar, despite the different degrees of wear of the samples (and the presence of wrinkles in the region of interest of the image of the second original sample-see image B in fig. 16 a). The statistical distribution of the wavelet coefficients obtained from the images of the two ink-jet printed replicas and the thermal sublimation replica (histograms C to E) is significantly different. The statistical distribution of wavelet coefficients (histogram F) derived from the image of the laser printed replica appears somewhat closer to the statistical distribution of the original sample. However, the dispersion of the histogram corresponding to the laser-printed replica is still smaller than that of the original sample. Furthermore, all histograms corresponding to the color replicas (histograms C to F) exhibit significantly different amplitudes and peak shapes compared to the histograms of the original samples (histograms a and B).

For illustrative purposes, fig. 17 shows a superposition of histograms corresponding to the first original sample (histogram a in fig. 16 b) and the laser printed color replica (histogram F in fig. 16 b).

FIGS. 18a and 18b are two bar graphs showing the variance σ calculated from the statistical distribution of wavelet coefficients obtained from images of substantially the same region of interest of eleven candidate documents, respectively²And an excess C, these candidate documents including five original samples (candidate documents 1 to 5) having different degrees of abrasion and six color copies (candidate documents 6 to 11) manufactured by inkjet printing, thermal sublimation, or color laser printing. In both cases, the variance σ²And excess C clearly indicate that it is possible to distinguish between real documents and counterfeits using these two statistical parameters as classification data.

For illustrative purposes, FIG. 19 uses the variance σ²And an excess C as a graphical representation of the corresponding feature space in (X, Y) coordinates in the feature space where the results derived from the candidate file can be located. A boundary line can be clearly drawn between a point corresponding to the original sample (located at the upper right corner of the feature space) and a point corresponding to the color reproduction (located at the lower left corner of the feature space).

FIG. 20 is a view of a feature space similar to that of FIG. 19, where the variance σ is²And the transition C is again used as the (X, Y) coordinate, and its display includes the original euro by processingAdditional candidate documents of banknotes. These results confirm the reliability and efficiency of the identification method according to the invention.

It should be realized that the method according to the invention therefore does not require that the selected region of interest is strictly one region of the candidate file and is the same region. In fact, deviations from one candidate file to another with respect to the actual position of the region of interest do not substantially affect the result. The method according to the invention is therefore also advantageous because it does not require precise identification and localization of the region of interest prior to signal processing. This greatly simplifies the entire authentication process and its implementation (particularly in ATM machines and the like), since it is only necessary to ensure that the selected region of interest more or less covers the region containing a sufficient typical number of intrinsic features (particularly intaglio features).

The authentication method described above may thus be summarized as shown in the flow chart of fig. 22, comprising the steps of:

acquiring a sample image (i.e. image c) of at least one region of interest r.o.i. of the surface of a candidate document to be authenticated⁰) The region of interest r.o.i encompasses at least a portion of the security feature;

by performing a decomposition of the sample image into at least one scale subspace containing high resolution details of the sample image (e.g. sub-image d)^m ₁，d^m ₂，d^m ₃Where m 1, 2,.., N, and N is the number of iterations performed) and extracting classification features (i.e., statistical parameters regarding the statistical distribution of spectral coefficients) from the scale subspace to digitally process the sample image c⁰(ii) a And

-obtaining an authenticity rating (or classification) of the candidate document based on the extracted classification features.

Fig. 21 schematically shows an implementation of a device for checking the authenticity of security documents, in particular banknotes, according to the method described above. The equipment bagComprises the following steps: an optical system 100 for obtaining a sample image (image c) of the region of interest r.o.i. on the candidate document 1 to be authenticated⁰) (ii) a And a digital signal processing unit (DSP)200 programmed to perform digital processing of the sample image. The DSP 200 may be particularly advantageously implemented as a Field Programmable Gate Array (FPGA) unit.

It will be appreciated that the invention described above may be applied to the simple detection of security features (particularly intaglio printed patterns) printed, coated or otherwise provided on security documents, particularly banknotes, which security features include characteristic visual features inherent to the process used to manufacture the security documents. As described above, classification features characterizing a security feature may be extracted by digitally processing a sample image of at least one region of interest of the surface of the candidate document (e.g., by performing one or more iterations of a multi-resolution analysis of the sample image), the region of interest being selected to include at least a portion of the security feature.

As explained above, the classification characteristic may be conveniently selected from the group consisting of arithmetic mean, variance (σ)²) Skewness, transition (C), and entropy of the statistical distribution of spectral coefficients representing fine-scale, high-resolution details of the sample image.

It will also be appreciated that the level of authenticity calculated according to the method described above may be optimised by designing the security feature to be printed, coated or otherwise provided on the security document so as to produce a feature response in a scale subspace or subspace comprising the high resolution details of the sample image being processed.

Such optimization may be achieved in particular by acting on security features comprising intaglio patterns, line offset patterns, relief patterns, light diffractive structures and/or combinations thereof. As shown for example in fig. 2b, a high density of such patterns, preferably gravure-printed straight or curved patterns, would be particularly desirable.

Various alterations and/or modifications may be made to the above-described embodiments without departing from the scope of the invention as defined in the appended claims.

For example, as already mentioned, although the authentication principle is preferably based on the processing of images containing (or supposed to contain) intaglio printed patterns, the invention is similarly applicable to the processing of images containing other security features including characteristic visual features inherent to the processes used to manufacture security documents, in particular line offset printed patterns, relief printed patterns, light diffractive structures and/or combinations thereof.

While the wavelet transform is discussed in the context of the above-described embodiments of the present invention, it should be recognized that this particular transform is considered a preferred transform within the scope of the present invention. However, other transformations are also possible, such as the so-called chirp transformation. From a general point of view, any suitable transform may be used as long as it is capable of performing a decomposition of the sample image into at least one scale subspace containing high resolution details of the sample image.

Furthermore, it will be appreciated that the above described method may be applied so that the sample image is decomposed into more than one scale subspace including high resolution details of the sample image at different scales. In this case, classification features may be extracted from each scale subspace in order to characterize the identified candidate document. In other words, the present invention is not limited to decomposing a sample image into only one scale subspace containing the high resolution details of the sample image.

In addition, although the process of statistical distribution of spectral coefficients is described as a method of extracting classification features to derive a level of authenticity of a candidate document to be authenticated, any other suitable process may be envisaged as long as such process is capable of separating and deriving features sufficient to represent the security features of an authentic security document.

Claims (31)

1. A method for checking the authenticity of a security document, wherein an authentic security document comprises a security feature (41-49; 30; 10; 51, 52) printed, coated or otherwise provided on the security document, said security feature comprising characteristic visual features inherent to the process used to manufacture said security document,

wherein the method comprises the steps of:

-obtaining a sample image (c) of at least one region of interest (r.o.i.) of the surface of the candidate document to be authenticated⁰) Of said interestSurrounds at least a portion of the security feature;

-performing the sample image (c)⁰) Decomposing to include the sample image (c)⁰) High resolution details (d)¹ ₁，d¹ ₂，d¹ ₃,..) and extracting classification features (σ) from the scale subspace²C..) to digitally process the sample image (C.,)⁰) (ii) a And

-based on the extracted classification features (σ)²C. -) obtain an authenticity rating of the candidate document.

2. The method of claim 1, wherein the step of digitally processing the sample image comprises:

-performing a registration of the sample image (c)⁰) To obtain a representation of said sample image (c)⁰) Of fine scale of (d)¹ ₁，d¹ ₂，d¹ ₃,..); and

-processing the spectral coefficients to extract the classification features (σ)²，C，...)。

3. The method of claim 2, wherein the processing the spectral coefficients comprises performing a processing of a statistical distribution of the spectral coefficients.

4. The method according to claim 3, wherein said processing of the statistical distribution of the spectral coefficients comprises calculating a variance (σ) selected from the group consisting of an arithmetic mean of the first moments in statistics, a variance of the second moments in statistics, a variance of the first moments in statistics, and a statistical distribution of the spectral coefficients²) -skewness, called third moment in statistics, -transition (C), called fourth moment in statistics, and-at least one statistical parameter of the group of entropy.

5. A method as claimed in any one of claims 2 to 4, wherein the transform is a wavelet transform.

6. The method of claim 5, wherein the wavelet transform is a Discrete Wavelet Transform (DWT).

7. The method of claim 6, wherein the Discrete Wavelet Transform (DWT) is selected from the group consisting of a Haar wavelet transform, a Daubechies wavelet transform, and a Pascal wavelet transform.

8. The method of any one of claims 1 to 4, wherein the sample image (c) is processed⁰) Is performed as a result of one or more iterations (N) of a multi-resolution analysis (MRA) of the sample image.

9. A method for checking the authenticity of a security document, wherein an authentic security document comprises a security feature (41-49; 30; 10; 51, 52) printed, coated or otherwise provided on the security document, said security feature comprising characteristic visual features inherent to the process used to manufacture said security document, said method comprising digitally processing a sample image (c.o.I) of at least one region of interest (R.o.I) of the surface of a candidate document to be authenticated⁰) The digital processing comprising the execution of said sample image (c)⁰) One or more iterations (N) of the multi-resolution analysis (MRA).

10. A method as claimed in any one of claims 1 to 4 and claim 9, comprising digitally processing a plurality of sample images corresponding to several regions of interest of the same candidate document.

11. The method of any one of claims 1 to 4 and claim 9, wherein at less than 600dpiResolution obtaining the sample image (c)⁰)。

12. The method of claim 11, wherein the resolution is 300 dpi.

13. The method according to any one of claims 1 to 4 and claim 9, wherein the security feature comprises an intaglio pattern (41-49), a line offset pattern (10), a relief pattern (51, 52), a light diffractive structure (30) and/or combinations thereof.

14. The method of any one of claims 1 to 4 and claim 9, wherein the security features comprise a rectilinear or curvilinear pattern of different widths, lengths and spacings.

15. The method of any one of claims 1 to 4 and claim 9, wherein the at least one region of interest (r.o.i) is selected to comprise a high density pattern.

16. The method of claim 15, wherein the pattern is a gravure printed straight or curved line pattern.

17. The method of claim 15, wherein the at least one region of interest (r.o.i) is selected to include a pattern of picture representations arranged on the candidate file.

18. A method according to any one of claims 1 to 4 and claim 9, wherein the security document is a banknote.

19. A digital signal processing unit (200) for processing a list of candidate documents (1) to be authenticated according to the method of any one of the preceding claimsSample image (c) of at least one region of interest (R.o.I) of a surface⁰) The digital signal processing unit (200) being programmed to perform a processing of the sample image (c)⁰) The digital processing of (2).

20. The digital signal processing unit (200) of claim 19, the digital signal processing unit (200) being implemented as a Field Programmable Gate Array (FPGA) unit.

21. Device for checking the authenticity of a security document according to the method of any one of claims 1 to 18, said device comprising means for obtaining a sample image (c) of a region of interest (r.o.i.)⁰) And an optical system (100) programmed to image (c) the sample⁰) A digital signal processing unit (200) that performs digital processing.

22. The apparatus of claim 21, wherein the digital signal processing unit (200) is implemented as a Field Programmable Gate Array (FPGA) unit.

23. A method for manufacturing a security document, the method comprising the step of designing a security feature to be printed, coated or otherwise provided on the security document, wherein the security feature is designed to optimize a level of authenticity calculated according to the method of any one of claims 1 to 8 by generating a feature response in at least one scale subspace.

24. The method according to claim 23, wherein the security feature comprises an intaglio pattern (41-49), a line offset pattern (10), a relief pattern (51, 52), a light diffractive structure (30) and/or combinations thereof.

25. A method as claimed in claim 23 or 24, wherein the security feature is designed to comprise a high density pattern.

26. The method of claim 25, wherein the pattern is a gravure printed straight or curved line pattern.

27. A method according to claim 23 or 24, wherein the security document is a banknote.

28. A method for detecting a security feature (41-49; 30; 10; 51, 52) printed, coated or otherwise provided on a security document, the security feature (41-49; 30; 10; 51, 52) comprising a characteristic visual feature inherent to a process for manufacturing the security document, the method comprising digitally processing a sample image (c) of at least one region of interest (R.o.I) of a surface of a candidate document⁰) Is selected to comprise the security feature (41-49; 30, of a nitrogen-containing gas; 10; 51, 52), the digital processing comprising performing the sample image (c)⁰) To extract a plurality of iterations (N) of a multi-resolution analysis (MRA) characterizing the security features (41-49; 30, of a nitrogen-containing gas; 10; 51, 52) of the classification features (σ)²，C，...)。

29. The method of claim 28, for detecting intaglio printed patterns (41-49).

30. Method according to claim 28 or 29, wherein said classification characteristic (σ)²C..) is selected from the group consisting of representing the sample image (C)⁰) Fine scale high resolution details of (d)¹ ₁，d¹ ₂，d¹ ₃,..) referred to as the arithmetic mean of the first moment in statistics, referred to as the variance (σ) of the second moment in statistics²) A skewness called third moment in statistics, a transition (C) called fourth moment in statistics, and a statistical parameter in the group of entropy.

31. A method according to claim 28 or 29, wherein the security document is a banknote.