EP2901366A1

EP2901366A1 - Method for detecting the reality of venous networks for the purposes of identifying individuals, and biometric recognition method

Info

Publication number: EP2901366A1
Application number: EP13766573.3A
Authority: EP
Inventors: William Ketchantang; Sylvaine Picard
Original assignee: Morpho SA
Current assignee: Idemia Identity and Security France SAS
Priority date: 2012-09-28
Filing date: 2013-09-27
Publication date: 2015-08-05
Also published as: JP2016500860A; JP6181763B2; BR112015006933A2; FR2996331A1; US9460335B2; FR2996331B1; WO2014049149A1; US20150261993A1

Abstract

The invention relates to a method for detecting a venous network in a portion of a human body, including the steps of: capturing an infrared image of said portion of a human body in order to reveal a venous network of said portion of a human body, carrying out a contrast analysis of the infrared image to determine contrast characteristics of the infrared image, determining if the portion of a human body is real from the contrast characteristics determined in the infrared image and the contrast characteristics determined from reference human body portions. The invention also relates to a biometric recognition method using a preliminary phase for detecting the reality of a portion of a human body.

Description

METHOD FOR DETECTING THE REALITY OF VENOUS NETWORKS FOR IDENTIFICATION OF INDIVIDUALS AND BIOMETRIC RECOGNITION METHOD

The present invention relates to a method of detecting a venous network on a body part of an individual for the purpose of identifying said individual. The invention also relates to a method of biometric recognition of an individual.

Automatic biometric recognition methods are increasingly used for the identification of individuals for example in the context of border control, access control to secure areas such as airport boarding areas or others, or data access controls.

A biometric recognition method generally comprises the following steps:

capture biometric characteristics on the body part of an individual,

compare the biometric characteristics to stored biometric characteristics,

- determine from the comparison whether the individual is recognized or not.

A biometric recognition method is implemented by means of an automated biometric recognition system comprising biometric signature capture means forming a signature, biometric signature storage means forming signatures and means for comparing captured biometric characteristics memorized biometric characteristics. The capture means are often image capture means. The storage or storage means are for example a memory (integrated circuit, barcode or other) which is embedded in a passport for storing biometric characteristics of the passport holder or a memory of a computer unit containing a database associating the biometric characteristics of individuals with identification data of these individuals. The comparison means are computer units incorporating calculation means for executing algorithms for processing and comparing the biometric characteristics.

The biometric characteristics of the hand and more particularly the fingers of individuals are very often used for the implementation of these methods. It is thus known to perform a biometric recognition from fingerprints or venous networks of the fingers. In the latter case, the capture means consist of an image sensor.

In the context of an automatic biometric recognition process, there is no operator to monitor the capture so that it is possible for a fraudster to attempt to deceive the system by presenting to the capture means a representation of biometric characteristics of an enlisted person, that is to say a duly identified person whose biometric characteristics have been memorized to allow recognition by the system. This representation may take the form of a fake finger provided with biometric characteristics or a simple image of a venous network.

Automated biometric recognition systems are thus provided with fraud detection means for this purpose. These means, generally of the hardware type, aim, for example, to detect fraud by:

detecting a variation of blood flow in the venous network as a function of finger pressure on the sensor; or - analyzing by interferometry a wave scattered by the finger; or even

- studying the differences in absorption by the finger of two light beams of different wavelengths.

These systems require a suitable technical architecture that makes them relatively expensive.

It is thus understood that the biometric identification of an individual supposes to answer two distinct problems:

- to determine if the part of body presented is indeed a part of alive body;

extracting a biometric signature from the body part to compare it with stored biometric signatures.

Indeed, if a fraudster manages to reproduce an artificial biometric signature, he can satisfy the identification.

It follows that the determination of the living character of the body part can not result from the mere success in the comparison of the biometric signatures.

In fact, the biometric recognition algorithms are arranged only to extract digital signatures from a medium, whatever its nature, and compare them to stored biometric signatures. On the contrary, the fraud detection algorithms are arranged to determine the nature of the support presented without extracting the biometric signature. This is because the characteristics that will reveal the living character of the medium and those that will allow recognition are not the same or are not exploited in the same way.

An object of the invention is to provide a simple means for detecting fraud.

For this purpose, it is provided, according to the invention, a method of detecting a venous network in a living human body portion, comprising the steps of:

capturing an infrared image of said human body part to bring out a venous network of said human body part,

- perform a contrast analysis of the infrared image to determine the contrast characteristics of the infrared image,

determining whether the infrared image has been captured on a real human body part from the contrast characteristics determined in the infrared image and from contrast characteristics determined from reference human body parts.

It appears indeed that it is very difficult to make an artificial representation of a venous network whose infrared image would present during a contrast analysis the same characteristics as a venous network image captured directly on a human body part living. Contrast analysis of the infrared image of the venous network thus allows detection of fraud. Fraud detection is also effective with infrared images captured in a single wavelength or range of wavelengths.

According to a particular mode of implementation, the method comprises a configuration phase and a detection phase.

The configuration phase includes the steps of:

capturing infrared images of human body parts and venous network representations;

isolating in each image blocks of pixels and performing a contrast analysis to determine for each block a directional gradient for at least two resolutions of the image; - determine average energies of the directional gradients;

determining by learning the average energy thresholds for discriminating a venous network image captured directly on a human body part of a venous network image captured on a representation of a venous network.

The detection phase comprises the steps of:

capturing at least one infrared image of a venous network;

isolating in the image blocks of pixels and performing a contrast analysis to determine for each block a directional gradient for at least two resolutions of the image;

- determine average energies of the directional gradients;

- Compare these mean energies with the thresholds to verify that the venous network image was captured directly on a part of the human body.

This mode of implementation is particularly effective. It is also relatively simple: the detection of fraud is carried out without taking into account geometric information (position, orientation) relative to the gradient, but only the average values of the amplitudes of the gradients. As a result, the comparison is made on a limited number of data, and is therefore time consuming and resource intensive. In addition, the directional gradients are not directly compared to directional gradients stored in relation to particular users, but mean energies from these directional gradients are compared with thresholds. This is more an observation than a comparison, in the sense of "matching", as usually used in biometric recognition. The invention also relates to a biometric recognition method comprising a phase of detecting the reality of a human body part and a biometric recognition phase comprising the following steps:

capture biometric characteristics on the body part of an individual,

compare the biometric characteristics with stored biometric characteristics,

determine from the comparison whether the individual is recognized or not,

the detection phase being performed by the implementation of the detection method according to the invention and the recognition phase being performed only if the human body part is determined as real at the end of the detection phase.

Thus, the recognition phase is started only if the detection phase is successful.

Other features and advantages of the invention will emerge on reading the following description of a particular non-limiting embodiment of the invention.

Reference will be made to the attached single figure schematically showing a system for implementing the method of the invention.

With reference to this figure, the invention relates to a method for detecting a venous network in a human body part, here the hand and more particularly a finger 1 of the hand.

The method is implemented by means of an automated system comprising a detection device 10, known per se, comprising an infrared illuminator 11 and an infrared image sensor 12 which are arranged opposite a positioning area finger 1 of the user. The positioning area is at the Once in the beam of the infrared illuminator 11 and in the field of the infrared image sensor 12. The positioning area can be materialized by a window against which the finger 1 can be applied and behind which are mounted the infrared illuminator 11 and the infrared image sensor 12, or the positioning area may be materialized by guides allowing the user to place his finger 1 in the beam of the infrared illuminator 11 and in the field of the image sensor infrared 12. The infrared frequency is chosen because it is strongly absorbed by hemoglobin and weakly absorbed by the tissues surrounding the venous network in a part of the human body: the venous network therefore exits with respect to the tissues surrounding it.

The infrared illuminator 11 and the infrared image sensor 12 are connected to a computer processing unit 20 known per se.

The processing computer unit 20 comprises a memory 21 containing a database of biometric characteristics, a program for managing the operation of the system, a program for processing the captured images to extract biometric characteristics and a comparison program for the biometric characteristics. extracted to the biometric characteristics of the database. In the embodiment described here, these different programs are portions of a single program arranged to implement the method of the invention.

The computing processing unit 20 also comprises a computing unit 22 arranged to execute the program in question and to control the entire system. The computer processing unit 20 also includes an input / output console allowing an operator to intervene on the system. The method of the invention comprises the steps of:

capturing an infrared image of the finger placed by the user in the capture area to bring out a venous network of said finger,

determining whether the infrared image has been captured on a real finger from the contrast characteristics determined in the infrared image and contrast characteristics determined from reference fingers.

In the embodiment described here, the method of the invention comprises a configuration phase and a detection phase.

The configuration phase includes the steps of:

isolating in each image blocks of pixels and performing a contrast analysis to determine for each block a directional gradient for at least two resolutions of the image;

- determine average energies of the directional gradients;

It is therefore necessary in this configuration phase to have representations of venous network such as those that could be used by a fraudster. These representations may have the form of flat images, images wound on a cylindrical support, artificial reproductions of a human finger ...

The detection phase comprises the steps of: capturing at least one infrared image of a venous network;

- determine average energies of the directional gradients;

In both phases, the determination of the directional gradients comprises the steps of:

isolating in each image blocks of a first number of pixels and performing a contrast analysis in each of the blocks to determine a directional gradient in each block of the first number of pixels;

isolating in each image blocks of a second number of pixels greater than the first number of pixels;

performing a contrast analysis in each of the blocks to determine a directional gradient in each block of the second number of pixels.

All pixels in the image are copied during contrast analysis.

It is understood that, in the direction of the vein, the gradient is lower than in a direction transverse to the direction of the vein. The estimation of the direction of the gradients is preferably carried out by a maximization of the directional derivatives according to four directions (we retain the largest amplitude among the four directions of derivation).

There are veins of different thicknesses and contrasts that vary with the surrounding tissues. Contrast analysis keeps this information.

In an infrared image of venous network captured on a real finger, the finger is not uniform in that the image has darker areas. In the center of the finger, the inter-phalanges junction diffusing more in the infrared makes it possible to accentuate the luminosity, and thus to accentuate the contrast between these zones and the rest of the finger. Contrast analysis provides information on both the pattern of the venous network and the non-uniformity of the image. Performing a contrast analysis in two different resolutions makes it possible to increase the number of information obtained both on the venous network and on the nonuniformity of the light scattering in the real finger. The higher this number of information, the more reliable the discrimination between the image of a real venous network and the image of a representation of a venous network.

The second number of pixels (that of the largest blocks) is preferably equal to the square of the first number of pixels (that of the smaller blocks). The first number of pixels is advantageously equal to 9 (well suited for the detection of high contrast veins) and the second number of pixels is advantageously equal to 81 (well suited for the detection of large veins and low contrast).

Preferably, the blocks are superimposed to cover all the images. Thus, each block is shifted by one pixel relative to its neighbors.

Directional gradients are determined in at least four directions (0 °, 90 °, 180 °, 270 °) by one of the following methods:

- subtraction of gray levels of pixels; or

- SOBEL type filtering; or

- CANNY DERICHE type filtering.

Subtracting gray levels from pixels is the easiest way to determine a directional gradient.

The comparison is preferably made by firstly determining vectors from the mean energies of the directional gradients.

During the configuration phase, it will be possible for example to use an SVM (Support Vector Machine) learning machine to determine a separation surface of the average energy vectors corresponding to the venous network images captured directly on human fingers and mean energy vectors of venous network images captured on venous network representations. The separation surface is representative of the threshold values for discriminating venous network images captured directly on human fingers and venous network images captured on venous network representations.

During the detection phase, the average energy vectors of the captured image are projected onto the separation surface so as to determine according to a sign of the projection if the image has been captured directly on a human finger or on a venous network representation.

Of course, the invention is not limited to the embodiments described but encompasses any variant within the scope of the invention as defined by the claims.

In particular, the determination of gradients directionals can be performed differently as described above.

The number of pixels of the blocks may be different and for example be even (in this case the determination of the directional gradients will require to carry out in a known manner in itself an interpolation).

Directional gradients can be calculated for only one resolution of the image.

Claims

A method of detecting a venous network in a human body portion, comprising the steps of:

determining whether the human body part is real from the contrast characteristics determined in the infrared image and contrast characteristics determined from reference human body parts.

2. Method according to claim 1, comprising a configuration phase and a detection phase, the configuration phase comprising the steps of:

isolating in each image blocks of pixels and performing a contrast analysis to determine for each block a directional gradient;

- determine average energies of the directional gradients;

determining, by learning, average energy thresholds for discriminating a venous network image captured directly on a human body part of a venous network image captured on a representation of a venous network;

the detection phase comprising the steps of:

capturing at least one infrared image of a venous network;

isolate in the image blocks of pixels and perform a contrast analysis to determine for each block a directional gradient;

- determine average energies of the directional gradients;

The method of claim 2, wherein the directional gradients are determined for at least two resolutions of the image.

The method of claim 2, comprising the steps of:

The method of claim 4, wherein the second number of pixels is equal to the square of the first number of pixels.

The method of claim 5, wherein the first number of pixels is 9.

The method of claim 2, wherein the blocks overlap to cover all the images.

The method of claim 2 wherein:

vectors are determined from the mean energies of directional gradients;

- during the configuration phase, an SVM type machine is used to determine a mean energy vector separation surface of directional gradients corresponding to venous network images captured directly on a portion of human body and mean energy vectors of directional gradients of venous network images captured on venous network representations;

during the detection phase, the average energy vectors of the captured image are projected onto the separation surface so as to determine according to a sign of the projection if the image has been captured directly on a part of human body or on a representation of venous network.

The method of claim 2, wherein the directional gradients are determined by:

- subtraction of gray levels of pixels; or

- SOBEL type filtering; or

- CANNY-DERICHE type filtering.

Biometric recognition method comprising a phase of detecting the reality of a human body part, and a biometric recognition phase comprising the following steps:

capture biometric characteristics on the body part of an individual,

compare the biometric characteristics to stored biometric characteristics,

determine from the comparison whether the individual is recognized or not,

the detection phase being performed by the implementation of the method according to any one of claims 1 to 9 and the recognition phase being performed only if the human body part is determined as real at the end of the phase detection.