FR2950455A1 - METHOD FOR DETECTING FALSE FINGER FOR FINGERPRINT ACQUISITION SOFTWARE - Google Patents

Abstract

The invention provides a false finger detection method for improved performance fingerprint acquisition software. To this end, the method according to the invention claims the implementation of a STAT static analysis step comprising calculating a GLRL matrix of a finger IMG image. Optionally, but preferably, the invention further proposes to combine a STAT static analysis and a DYN dynamic analysis to optimize the false finger detection capability.

Description

Fingerprint detection method for fingerprint acquisition software.

The present invention relates to a new method enabling a fingerprint acquisition software to detect possible "false fingers". By implementing an unprecedented image analysis method in this field, the method according to the invention makes it possible to improve the performance of the software for acquiring fingerprints.

The acquisition and recognition of fingerprints has become indispensable in the context of the control of the identity of individuals. Current fingerprint acquisition systems include a hardware component and a software component. The hardware component is essentially based on the implementation of a sensor for acquiring an image of the finger whose fingerprint is to be acquired. The software component includes a human-machine interface allowing the acquisition of information, the recording of personal data, the display of information for the operator and the subject. Moreover, currently, a goal of fingerprint acquisition software is increasingly to detect possible fraud attempts of these systems, through the use of false fingers, such as molded fingers, fingers latex ... etc. These frauds can be motivated by a desire to hide one's identity so as not to be recognized, to impersonate someone, or to try to create a false identity for example. To ensure the efficiency of the fingerprint acquisition stations, it is important to develop systems for detecting these attempts of fraud.

The invention lies in this context of seeking more effective solutions 3o to detect fake fingers, through a software approach to the problem.

The state of the art includes a number of technical solutions for detecting false fingers. In particular, several technologies based on a material approach to the question have been developed, such as that described in US Patent 7,415,139; they are most often to determine whether the finger whose image is acquired is alive, through temperature measurements, blood flow, or by the detection of sweating, for example. Other technologies are based on an analysis of the image of the fingerprint; such techniques are for example disclosed in publications such as R. DERAKHSHANI ET AL, "Determination of vitality from a non-invasive biomedical measurement for use in fingerprint scanners", in Pattern Recognition, v 36, No. 2, p 383-96, Feb. 2003, or in S. NIKAM ET AL, "Wavelet energy signature and GLCM features based fingerprint anti-spoofing", for Sixth International Conference on Wavelet Analysis and Pattern Recognition (ICWAPR-2008), Hong Kong, Aug. 30-31. 2008, to give an example proposing a wavelet analysis of the acquired image.

The disadvantage of the known techniques lies mainly in their lack of efficiency. An object of the invention is in particular to overcome this lack of efficiency in dealing with the problem of the detection of false fingers, in the context of a fingerprint acquisition software, along two axes: the first axis consists of introduce a new criterion for analyzing a finger image; the second axis, making it possible to further improve performance in the context of a preferred embodiment of the invention, lies in the idea of combining a static image analysis and a dynamic analysis of a stream of images in order to remove a maximum of ambiguities in the context of a false finger detection method.

For this purpose, the subject of the invention is a method for detecting fingerprints for fingerprint acquisition software comprising a step of acquiring the image of a finger, characterized in that it comprises a step static analysis of said finger image, said static STAT analysis comprising: • calculating any of the gray-level chain length matrices, called GLRL matrices for Gray Level Run Length according to the dedicated acronym corresponding to said finger image; Constructing a first vector characteristic of the finger image comprising at least partially said GLRL matrix; The implementation of static analysis methods of said first characteristic vector.

In one embodiment of the invention, the static analysis methods comprise at least one of the following methods: a method for analyzing the periodicity of the pores of the skin; • a method of analysis of the gray level co-occurrence matrix, called GLCM for Gray Level Co-occurrence Matrix according to the dedicated acronym, associated with the finger image.

Advantageously, the static analysis step may be followed by the implementation of a principal component analysis method, called the ACP method. Advantageously, the static analysis step may be followed by the implementation of a supervised learning method from a learning database applied to the first characteristic vector. Advantageously, the static analysis step may be furthermore followed by the implementation of a kernel analysis applied to the first characteristic vector of the finger image based on said supervised learning method. Advantageously, the kernel analysis further comprises a classification step. Advantageously, the method according to the invention can determine, according to the results of the static analysis methods, a first score corresponding to a confidence index linked to the probability that the finger at the origin of the image analyzed is true. , respectively false.

Advantageously, the method according to the invention may also comprise an operator alerting step in the case where the first score is less than a first predetermined threshold.

Advantageously, the method according to the invention may furthermore comprise a step of acquiring a stream of finger images comprising a plurality of images of the finger, as well as a stage of dynamic analysis of said flow of images resulting in to the construction of a second vector characteristic of the image flow.

Advantageously, the dynamic analysis step is followed by a linear combination of the measurements constituting the second characteristic vector. Advantageously, the dynamic analysis step is capable of allowing the detection of a sweating phenomenon of the finger.

Advantageously, the dynamic analysis step comprises the implementation of one or more dynamic analysis methods among the following methods: an analysis of the temporal evolution on a dimension of the peak lines of the images of the stream of finger images; • a wavelet analysis of the image flow of the finger.

Advantageously, the image stream is stored on a computer medium for subsequent analysis. Advantageously, the dynamic analysis step may be followed by the implementation of a second supervised learning method from a learning database applied to the second characteristic vector. Advantageously, the dynamic analysis step may further be followed by the implementation of a kernel analysis applied to the second characteristic vector based on said supervised learning method. Advantageously, the method according to the invention can determine, according to the results of dynamic analysis methods, a second score corresponding to a confidence index related to the probability that the finger at the origin of the image analyzed is true.

Advantageously, the method according to the invention may also include an operator alerting step in the case where the second score is less than a second predetermined threshold.

Advantageously, the static analysis step and the dynamic analysis step respectively lead to the determination of first and second scores, and in that said method comprises a step of merging said first and second scores into to determine an improved confidence index related to the probability that the finger at the origin of the analyzed image is true. Advantageously, the method according to the invention may also comprise a step of alerting the operator in the case where the score is less than a second predetermined threshold.

The invention also resides in a fingerprint acquisition station comprising: a sensor for acquiring the image of one or more fingers; A computer assembly comprising a first display screen of a first human-machine interface for the operator of the acquisition station and a computer implementing software, the software implementing the previously described method.

Advantageously, said fingerprint acquisition station, further comprising a sensor for acquiring a stream of images of one or more fingers. Advantageously, said fingerprint acquisition station further comprises a second human-machine interface for the subject of which said acquisition station acquires the image of one or more fingers, said second human-machine interface can be displayed on the first display screen or on a second display screen.

Other characteristics and advantages of the invention will become apparent with the aid of the following description made with reference to the appended drawing which represents: FIG. 1: the block diagram of the principle of the invention according to its preferred embodiment.

Figure 1 shows a block diagram of the false finger detection method according to the invention, according to its preferred embodiment, incorporating certain optional features.

The method according to the invention is implemented from the moment when an IMG image of one or more fingers has been acquired by any optical sensor. The method includes a first STAT static analysis step of the IMG finger image. Optionally, the method according to the invention may comprise, in parallel, a dynamic analysis step DYN of the IMG image. Preferably, the dynamic analysis step DYN is performed only if the STAT static analysis step makes it possible to determine that it is necessary, or at least recommended, to perform this dynamic analysis step DYN. to improve the quality of the result obtained.

The STAT static analysis step results in the construction of a vector characterizing the IMG finger image. According to the invention, the STAT static analysis step comprises, for this purpose, a step of calculating a GLRL matrix, for Gray Level Run Length according to the consecrated acronym, corresponding to the IMG finger image. The calculation of a GLRL matrix for performing the analysis of an image is as such known. Thus, F. ALBREGTSEN, "Statistical Texture Measures Computed From Gray Level Run Length Matrices", for Image Processing Laboratory, Department of Informatics, University of Oslo, Nov. 1995, refers to it. In the field of false finger detection for fingerprint acquisition software, this practice is unknown. The calculation of any GLRL matrix associated with the IMG finger image makes it possible to have a precise characterization of the IMG finger image with a minimum of information. From this GLRL matrix, and possibly other results from different static analysis methods, the method according to the invention provides the construction of a vector characteristic of the IMG finger image from a set M1 measurements from the STAT static analysis step. According to the invention, the STAT static analysis step therefore comprises at least the calculation of a GLRL matrix associated with the IMG finger image. The analysis of the GLRL matrix gives information on the texture of the skin of the finger. Preferably, the calculation of said GLRL matrix will be combined with other static analysis methods such as: an analysis of the periodicity of the pores of the skin by the technique, known to those skilled in the art, of the transformations of Fourier called FFT, for Fast Fourier Transform according to the acronym, applied to the signal lines of the crest of the fingerprint detected on the image IMG; An analysis of the roughness by a wavelet analysis method of the signal of the valleys of the fingerprint detected on the IMG image; • an analysis of the texture of the skin of the finger, this time using the analysis of the gray level co-occurrence matrix, called GLCM for Gray Level Cooccurrence Matrix according to the acronym English.

The method according to the invention can comprise, after this STAT static analysis step resulting in the construction of a measurement vector M1 characteristic of the IMG finger image, a processing step 25 and CLASS classification of these measurements M1. by a statistical approach, such as a principal component analysis method, called the ACP method, or a known fuzzy logic technique. This step, particularly in the case where the STAT static analysis step comprises the analysis of different criteria, preferably corresponds to a so-called "core" analysis comprising a phase based on a supervised learning method from a learning database, and a classification phase itself. In any case, the method according to the invention then makes it possible to calculate a first score, or index, SCOI, expressing a confidence index relative to the probability that the finger whose IMG image is in progress. analyze is a real finger. This SCOI score is compared with a first predetermined threshold Si. If the first SCOI score is greater than the first predetermined threshold S1, a margin can be predicted, then the confidence index is sufficient and the prospective operator can be warned.

According to a first embodiment, the false finger detection method according to the invention has now come to an end. Optionally, it may nevertheless still include a dynamic analysis step DYN, as described below, as a verification.

If the first score SCOI is lower than the first predetermined threshold SI, a margin can be expected, then the index of confidence is insufficient and the prospective operator can be warned that it is likely that the finger whose image IMG is being analyzed is wrong. The operator warning may take the form of an ALT warning message. This warning or this ALT warning can, according to the simplest embodiment, occur without a step DYN dynamic analysis of the IMG image is implemented, even if this is not shown in the figure According to the preferred embodiment of the method according to the invention, when the first SCOI score is below the predetermined threshold SI, a margin that can be provided, said method comprises an additional dynamic analysis step DYN. However, this dynamic analysis step DYN can be carried out systematically, for example in parallel with the STAT static analysis step. The dynamic analysis DYN is based on the evolution over time of a plurality of IMG images acquired over a certain duration. The sensor then acquires an IMG image stream. Typically, such a sensor can acquire two IMG images spaced two or three seconds apart. This dynamic analysis step DYN can comprise one or a combination of dynamic analysis methods, among the following methods: • a temporal analysis of the signal on a dimension of the fingerprint peak lines detected on the flow of IMG images; A wavelet analysis of the image flow and the calculation of the energy ratio between the initial image, or first image of the IMG image stream, and the final image, or last image of the IMG image stream.

The dynamic analysis step DYN results in the construction of a measurement vector M2 characteristic of the IMG image flow. In the method according to the invention, the dynamic analysis step DYN can make it possible to calculate at this time a second score, or index, SCO2, expressing a confidence index relative to the probability that the finger whose IMG image is being analyzed is a real finger. In FIG. 1, and in accordance with a preferred embodiment of the invention, the second SCO2 score is calculated after a COMB linear combination step of the M2 vector measurements has been performed. To improve the determination of the second SCO2 score, this COMB step may be followed by a learning phase from a learning database. Also, as in the case of the STAT static analysis step, a processing of the M2 measurement vector can be performed using a "kernel" analysis based on said learning method. Once determined, the second score SCO2 is compared with a second predetermined threshold S2. If the second score SCO2 is greater than the second predetermined threshold S2, a margin can be expected, then the confidence index is sufficient and the prospective operator can be notified. In the opposite case, ie if the second SCO2 score is lower than the second predetermined threshold S2, a margin can be predicted, then the confidence index is insufficient and the prospective operator can be warned that it is likely that the finger whose IMG image and an IMG image stream are being analyzed is false. The operator warning may take the form of an ALT warning message.

According to the preferred embodiment of the method according to the invention, in which the dynamic analysis step DYN is performed only if the first SCOI score is lower than the first predetermined single IF, a FUS fusion of the first SCOI and SCO2 scores are performed before comparing the result to the second threshold S2. It is then the result of the FUS fusion that is compared to the second predetermined threshold S2. If FUS fusion of the first SCOI and second SCO2 scores is greater than the second predetermined threshold S2, a margin can be expected, then the confidence index is sufficient and the prospective operator can be notified. In the opposite case, that is, if the FUS fusion of the first SCOI and the second SCO2 score is less than the second predetermined threshold S2, a margin can be provided, then the confidence index is insufficient and the prospective operator can be warned that it is likely that the finger whose IMG image and an IMG image stream are being analyzed is wrong. The operator warning may take the form of an ALT warning message.

Furthermore, in a particular embodiment of the invention, the IMG image stream can be stored on a computer medium for subsequent analysis, so as not to slow down the overall process of acquiring fingerprints. .

It should be noted that the present invention also covers a fingerprint acquisition station capable of implementing the method described above. Such an acquisition station comprises a sensor configured to perform the acquisition of the image of one or more fingers and, optionally, the acquisition of a stream of images of one or more fingers. The acquisition station according to the invention further comprises a computer unit equipped with software implementing the method according to the invention. In addition, said acquisition station may comprise a first screen having a human-machine interface for the operator of the station for the purpose of displaying information or entering information. The station may also include a second screen having a human-machine interface to the subject of which one or more fingers are being acquired, for the purpose of displaying information essentially.

To summarize, the invention has the main advantage of providing a false finger detection method for enhanced performance fingerprint acquisition software. To this end, the method according to the invention claims the implementation of a static analysis step comprising calculating a GLRL matrix of a finger image. Optionally, but preferably, the invention further proposes to combine static analysis and dynamic analysis to optimize false finger detection capability.

Claims (20)

REVENDICATIONS1. Fingerprint detection method for fingerprint acquisition software comprising a step of acquiring the image (IMG) of a finger, characterized in that it comprises a static analysis step (STAT) of said finger image (IMG), said static STAT analysis comprising: • calculating any of the gray-level chain length matrices, called GLRL matrices for Gray Level Run Length according to the dedicated acronym, corresponding to said finger image (IMG); Constructing a first characteristic vector (M1) of the finger image (IMG) comprising at least partially said GLRL matrix; The implementation of static analysis methods of said first characteristic vector (M1). 2. Method according to claim 1, characterized in that said static analysis methods (STAT) comprise at least one of the following methods: a method for analyzing the periodicity of the pores of the skin; • a method of analysis of the gray level co-occurrence matrix, called GLCM for Gray Level Cooccurrence Matrix according to the acronym, associated with the finger image. 25 3. Method according to any one of claims 1 to 2, characterized in that the static analysis step is followed by the implementation of a principal component analysis method, called ACP method. 4. Method according to any one of claims 1 to 3, characterized in that the static analysis step (STAT) is followed by the implementation of a supervised learning method from a base 15 30data of learning applied to the first feature vector (MI). 5. Method according to claim 4, characterized in that the static analysis step is followed by the implementation of a core analysis (CLASS) applied to the first characteristic vector (MI) of the image (IMG). fingerprint based on said supervised learning method. 6. Method according to claim 5, characterized in that the nucleus analysis (CLASS) further comprises a classification step. 7. Method according to any one of claims 1 to 6, characterized in that it determines, according to the results of static analysis methods (STAT), a first score (SCOI) corresponding to a confidence index related to the probability that the finger at the origin of the analyzed image is true, respectively false. 8. The method of claim 7, characterized in that it comprises an alert step (ALT) of the operator in the case where the first score (SCOI) is less than a first predetermined threshold (SI). 9. Method according to any one of claims 1 to 8, characterized in that it further comprises a finger image acquisition step (IMG) comprising a plurality of images of the finger, and a step of dynamic analysis (DYN) of said image flow (IMG) resulting in the construction of a second characteristic vector (M2) of the image flow (IMG). 10. Method according to claim 9, characterized in that the dynamic analysis step (DYN) is followed by a linear combination (COMB) of the measurements constituting the second characteristic vector (M2). 11. Method according to any one of claims 9 to 10, characterized in that the dynamic analysis step (DYN) is adapted to allow the detection of a sweating phenomenon of the finger. 12. Method according to any one of claims 9 to 11, characterized in that the dynamic analysis step (DYN) comprises the implementation of one or more dynamic analysis methods among the following methods: analyzing the temporal evolution over a dimension of the crest lines of the images of the image flow (IMG) of the finger; • a wavelet analysis of the image flow (IMG) of the finger. 13. A method according to any one of claims 9 to 12, characterized in that the image stream (IMG) is stored on a computer medium for subsequent analysis. 14. Method according to any one of claims 9 to 13, characterized in that the dynamic analysis step (DYN) is followed by the implementation of a second supervised learning method from a learning database applied to the second feature vector (M2). 15. Method according to claim 14, characterized in that the dynamic analysis step (DYN) is followed by the implementation of a kernel analysis applied to the second characteristic vector (M2) based on said method of analysis. supervised learning. 16. Method according to any one of claims 9 to 15, characterized in that it determines, according to the results of the dynamic analysis methods (DYN), a second score (SCO2) corresponding to a linked confidence index. to the probability that the finger at the origin of the analyzed image is true. 17. The method of claim 16, characterized in that it comprises an alert step (ALT) of the operator in the case where the second score (SCO2) is less than a second predetermined threshold (S2). 18. Method according to any one of claims 9 to 15, characterized in that the static analysis step (STAT) and the dynamic analysis step (DYN) respectively lead to the determination of a first one (SCOI). ) and a second (SCO2) scores, and in that said method comprises a step of fusing (FUS) said first (SCOI) and second (SCO2) scores to determine an improved confidence index related to the probability that the finger at the origin of the analyzed image is true. 19. The method of claim 18, characterized in that it comprises an operator ALT step in the case where the score is less than a second predetermined threshold (S2). 20. A fingerprint acquisition station comprising: a sensor for acquiring the image (IMG) of one or more fingers; A computer assembly comprising a first display screen of a first human-machine interface intended for the operator of the acquisition station and a computer implementing software, characterized in that the software implements the A method according to any one of claims 1 to 8. 23. A fingerprint acquisition station according to claim 20, further comprising a sensor for acquiring an image stream (IMG) of a or several fingers, characterized in that the software implements the method according to any one of claims 9 to 19. 24. Acquisition station according to one of claims 20 to 21, characterized in that it comprises in addition, a second human-machine interface for the subject of which said acquisition station acquires the image (IMG) of one or more fingers, said second human-machine interface being able to be displayed on the first display screen Or on a second display screen.5