FR3110010A1 - Process for acquiring a biometric trait of an individual, for authentication or identification of said individual - Google Patents

Abstract

The present invention relates to a method for acquiring a biometric trait of an individual, characterized in that it comprises the implementation by data processing means (11) of a terminal (1) of the following steps : Obtaining, from first optical acquisition means (14a) of the terminal (1), a first image of said individual representing said biometric trait on a background; Generation of a second image, representing said biometric feature of the first image on a bright background; Display of said second image on an interface (13) of said terminal (1) so as to illuminate the individual with said luminous background. Abstract figure: Fig. 1

Description

Method for acquiring a biometric trait of an individual, with a view to authentication or identification of said individual

GENERAL TECHNICAL AREA

The present invention relates to the field of authentication and biometric identification, in particular by face or iris recognition.

STATE OF THE ART

Biometric access control terminals are known, in particular based on optical recognition: an authorized user places a biometric feature (his face, his iris, etc.), in front of the terminal, the latter is recognized, and an access for example is unlocked.

Today this technology is embedded in personal mobile terminals of the smartphone type: by means of a front camera of the terminal, the biometric trait of the user can be recognized, in particular to unlock the terminal or access a sensitive service ( for example a banking application).

The difficulty that arises is acquisition in a dark environment (at night, indoors, etc.). Indeed, the mobile terminal being a nomadic device, it must be able to be used everywhere and not necessarily in a standardized environment. It is known to “adjust” the camera to adapt the exposure according to the luminosity observed, but in too dark an environment, overexposure causes a noisy image which makes it difficult to recognize and/or detect fraud.

The most natural solution is to illuminate the user's face with a light source. However, while most smartphones have a flash on the back side, few have on the front side, so a more universal method is needed.

It has thus been proposed to use the terminal screen as a light source, firstly by displaying an entirely white screen for a short time. However, this solution is not very ergonomic, because it cuts off the video feedback, that is to say the possibility for the user to observe the acquired image so as to position his face correctly. It is possible to chain acquisition phases where the screen is white and verification phases where the acquired image is displayed, but this remains impractical and even aggressive for the eyes.

To have continuous video feedback while benefiting from screen lighting, it was then proposed to display a white “band” or “outline”, superimposed on the video feedback like a frame. However, the brightness provided is limited because the area of the white zone is small, and in addition the video feedback is small or cropped, which prevents the user from using it properly.

It would therefore be desirable to have a new solution that is simple, reliable and effective whatever the lighting conditions, for the acquisition of a biometric trait, with a view to authentication or biometric identification.

PRESENTATION OF THE INVENTION

According to a first aspect, the present invention relates to a method for acquiring a biometric trait of an individual, characterized in that it comprises the implementation by data processing means of a terminal of the following steps:

1. Obtaining, from first optical acquisition means of the terminal, a first image of said individual representing said biometric trait on a background;
2. Generation, of a second image, representing said biometric feature of the first image on a luminous background;
3. Display of said second image on an interface of said terminal so as to illuminate the individual with said luminous background.

According to other advantageous and non-limiting characteristics:

The method includes a new occurrence of step (a) while the second image is displayed on the interface.

The second image corresponds to the first image in which the background has been at least partially replaced by said luminous background, in particular entirely replaced by said luminous background.

Step (b) includes the prior extraction of a region of interest from the first image containing the biometric feature, the second image being generated from the extracted region of interest.

Said extraction of a region of interest from the first image containing the biometric feature comprises the detection of the biometric feature of the individual in the first image, in particular by segmentation of the first image.

Step (b) further includes adapting the exposure of the first image to the detected biometric feature before extracting the region of interest.

The method further comprises a step (d) of authentication or identification of said individual on the basis of the detected biometric trait.

Step (d) comprises comparing the detected biometric trait with reference biometric data stored on data storage means.

Step (d) includes implementing access control based on the result of said biometric identification or authentication.

Said luminous background is a light plain background, preferably white.

Said biometric trait of the individual is selected from a face and an iris of the individual.

Said first image is a visible image.

According to a second aspect, the present invention relates to a terminal comprising data processing means configured to implement:

• Obtaining, from first optical acquisition means of the terminal, a first image of said individual representing said biometric trait on a background;
• The generation of a second image, representing said biometric feature of the first image on a luminous background;
• The display of said second image on an interface of said terminal so as to illuminate the individual with said luminous background.

According to a third and a fourth aspect, the invention proposes a computer program product comprising code instructions for the execution of a method according to the first aspect of acquiring a biometric trait of an individual; and a computer-readable storage medium on which a computer program product includes code instructions for performing a method according to the first aspect of acquiring a biometric trait of an individual.

PRESENTATION OF FIGURES

Other characteristics and advantages of the present invention will appear on reading the following description of a preferred embodiment. This description will be given with reference to the appended drawings in which:

• [Fig. 1] FIG. 1 generally represents a terminal for implementing the method for acquiring a biometric trait of an individual according to the invention;
• [Fig. 2] FIG. 2 schematically represents the steps of an embodiment of the method for acquiring a biometric trait of an individual according to the invention;
• [Fig. 3a] FIG. 3a represents an example of a first image used in the method according to the invention;
• [Fig. 3b] FIG. 3b represents an example of a second image used in the method according to the invention.

DETAILED DESCRIPTION

Architecture

With reference to the , a terminal 1 is proposed for the implementation of a method for acquiring a biometric trait of an individual. Preferably, this acquisition process is part of a process for authenticating or identifying said individual, that is to say aimed at determining or verifying the identity of the individual appearing in front of the terminal 1, to authorize access to this individual, if necessary. As we will see, this is typically facial biometrics (facial or iris recognition), in which the user must approach their face, but also remote fingerprint biometrics (digital or palm) in which the user approaches his hand.

The terminal 1 is typically a mobile terminal personal to an individual, such as for example a portable telephone or “smartphone”, an electronic tablet, a personal computer, etc.

In the remainder of this description, the preferred example of a smartphone type terminal will be taken, the present method then being generally used to authorize software access to the terminal 1 (for example unlocking the terminal 1, accessing personal data or an application, implement a transaction, etc.), but it will be noted that the present method remains usable in very many situations such as the prior authentication of an individual wishing to board an aircraft, the opening of a locked gate, etc.

The terminal 1 comprises data processing means 11, typically of the processor type, managing the operation of the terminal 1, and controlling its various components, usually in a box 10 protecting its various components.

Preferably, the terminal 1 comprises first optical acquisition means 14a, typically arranged so as to observe a space generally located "in front" of the terminal 1 (i.e. on the front face) and to acquire data, in particular images of a biometric trait such as an individual's face or iris. As will be seen, the first optical acquisition means 14a are advantageously arranged in the vicinity of an interface 13 of the terminal 1 (most often of the touch screen type), so as to be able to observe the face of an individual taking hand the terminal 1 and looking at the interface 13. Indeed, as we will see the present method implements a "video return" of the data acquired via the first optical acquisition means 14a, so that the user must to be able to simultaneously observe the interface 13 and be observed by the first optical acquisition means 14a. Typically, in the case of a smartphone, the first optical acquisition means 14a are arranged just above said interface 13.

As will be seen, the interface 13 plays the role of lighting means adapted to illuminate said space facing said first optical acquisition means 14a (that is to say that they will be able to illuminate the observable subjects by the optical acquisition means 14a, insofar as the user "looks" towards the interface 13 and therefore turns his face towards this interface 13). It is thus understood that the light emitted by the interface 13 is received and re-emitted by the subject to the terminal 1, which allows the optical acquisition means 14a to be able to acquire data of correct quality and increase the reliability of any subsequent biometric processing. Indeed, a face in the dark will be more difficult to recognize, for example. Also, we note that "spoofing" techniques (in French "usurpation" or "mystification") in which an individual tries to fraudulently deceive an access control terminal by means of accessories such as a mask or a prosthesis are easier to spot under adequate lighting.

Note that there may well be second optical acquisition means 14b which would observe another space (and which are not involved in the desired biometric operation): mobile terminals of the smartphone type generally have both cameras on the front face and on the rear face, and then the second optical acquisition means 14b are typically those located on the rear face, i.e. opposite to the first optical acquisition means 14a.

In the remainder of this description, the focus will be on the space located "facing" the optical acquisition means 14a, that is to say that located "opposite" the optical acquisition means 14a, which is therefore observable and in which one wishes to make biometric identification or authentication.

The first optical acquisition means 14a are typically sensors allowing the acquisition of a “radiation” image, that is to say a conventional image in which each pixel reflects the real appearance of the observed scene, ie where each pixel has a value corresponding to the amount of electromagnetic radiation received in a given part of the electromagnetic spectrum. Most often, said radiation picture is as seen in the a visible image (generally a color image - RGB type - for which the value of a pixel defines its color, but also a grayscale or even black and white image - for which the value of a pixel defines its luminosity) , ie the image as the human eye can see it (the electromagnetic spectrum concerned is the visible spectrum – band from 380 to 780 nm), but it can alternatively be an IR (infrared) image – for which the electromagnetic spectrum concerned is that of wavelength beyond 700 nm, in particular of the order of 700 to 2000 nm for the “near infrared” band, NIR), or even images linked to other parts of the spectrum.

It should be noted that the present method can use several radiation images in parallel, in particular in different parts of the electromagnetic spectrum, where appropriate respectively acquired via several different first optical acquisition means 14a. For example, a visible image and an IR image can be used. In the remainder of this description, the example of a visible image will be taken.

Finally, the data processing means 11 are often connected to data storage means 12 storing a reference biometric database, preferably images of faces (faces) or irises, so as to allow the comparison of a biometric feature of the individual appearing on the radiation image with the reference biometric data. The means 12 can be those of a remote server to which the terminal 1 is connected, but advantageously they are local means 12, i.e. included in the terminal 1 (in other words the terminal 1 includes the storage means 12 ), so as to avoid any transfer of biometric data over the network and limit the risk of interception or fraud.

Process

With reference to the , the present method, implemented by the data processing means 11 of the terminal 1, begins with a step (a) of obtaining an image on which appears a biometric feature of said individual, referred to as the first image. More precisely, the first image is an image of said individual representing (at least partially) said biometric trait, on a background.

Generally, this step includes the acquisition of data from these means 14a and the obtaining of the first image from the data acquired by the first optical acquisition means 14a.

By background of the image, we mean a generally "complex" background on which the biometric feature is superimposed, often the aspect of the place where the individual is located: typically the face of the individual appears "in front of this background. This background can be very dark, especially if the individual is in a small, poorly lit room or outside when the sun is down. For convenience, in the present description, the background is defined as "everything that is not the biometric trait" and therefore potentially including the body of the individual if the trait is his face (or even all the rest of the face if the feature is an eye).

The present method is distinguished in that it comprises a step (b) of generating a second image, representing said biometric feature of the first image on a luminous background. There represents an example second image corresponding to the first image of FIG. 3a.

This second image can then be displayed (instead of the first image) on the interface 13 as video feedback in a step (c), so as to illuminate the individual with said luminous background.

It is understood that the second image represents the entire biometric trait, and represents it in an identical manner to the first image, that is to say only modifies the background, which does not include useful information for a biometric application and can therefore be tampered with. So :

• Video feedback can be implemented permanently;
• The size or area of the biometric line is not reduced, which provides maximum comfort for the user during video feedback;
• We have a large surface that can be used as a light source for a next occurrence of step (a) so that we do not have an illumination problem.

More precisely, a new occurrence of step (a) is advantageously implemented while the second image is displayed on the interface 13 so that during the acquisition of a new first image the individual is illuminated by the luminous background of the second image, i.e. steps (a) to (c) are implemented cyclically.

Thus, preferably in step (a), a first stream of first images is obtained with in particular data acquisition at a predetermined frequency, for example 15Hz, each first image is processed on the fly so as to generate in step ( b) a second stream of second images, and displaying this second stream in step (c): this makes it possible to have steps (a) and (c) implemented simultaneously so as to have constant illumination of the individual and the first "improved" and thus good quality images. We naturally understand that the improvement of the first image slightly impacts the second image, and so on the next first image, but in a few images (and therefore a fraction of a second) we reach a stable state.

By "bright background" is meant any background having sufficient luminosity for the lighting requirements of the individual, generally a clear plain background, preferably white, but it is understood that it may be for example the same background plane but with a significantly enhanced luminosity (decorrelated from the biometric feature). As we will see, the quality of light produced by the luminous background can be dynamically adapted, in particular either by varying its color, or by adjusting the rate of the background which is replaced by the luminous background (advantageously the whole background), or even playing on the zoom to adjust the rate of the first image which is occupied by the biometric feature.

The idea of step (b) is to isolate the biometric trait (or at least a “region of interest” containing the biometric trait) from the first image and use it as a basis for the second image, i.e. that the second image corresponds to the first image in which the background has been at least partially replaced by said luminous background.

By region of interest is meant a semantically more interesting spatial zone containing the biometric trait, typically the whole body if the biometric trait is the face, or the whole face if the biometric trait is the iris.

Here, the replacement is preferentially "total", i.e. that said zone of interest coincides with the biometric line, but as explained it is understood that it is possible to keep zones which strictly speaking form part of the background, in particular the rest of the person as is the case in the example of figure 3b: if the biometric feature is only the face, the neck and the chest of the individual are technically part of the background, but for aesthetic issues it may be interesting not to represent the floating face alone in the second image and therefore to also keep the body (which in this example is the region of interest). For convenience, in the remainder of this description, the example will be taken in which the region of interest coincides with the biometric trait, but it is possible to remain a little less specific and have a region of interest going a little beyond the single biometric trait (in all cases the region of interest must contain the biometric trait).

Step (b) thus advantageously comprises two sub-parts, namely first of all the extraction of the biometric feature (or at least of the region of interest) from the first image, then the generation of the second image at from the extracted biometric trait.

To do this, one advantageously begins with the prior detection of the biometric trait of the individual in the first image with a view to locating it, and more precisely the so-called "semantic" segmentation of the first image, that is to say the separation of the biometric feature (or at least the region of interest) and the background by image analysis. More precisely, the semantic segmentation makes it possible to “clip” the biometric feature/the region of interest.

As explained, it is still possible to separate only a “region of interest” containing the biometric feature from the rest of the image, for example the whole person. In this case, the detection will always relate to the biometric trait, but then the segmentation will take into account the region of interest. More precisely, the segmentation can in practice distinguish several zones of the first image, including the biometric feature and parts of the region of interest, it is then sufficient to select the corresponding zones to reconstitute the region of interest.

Preferably, the detection of the biometric trait and the extraction of the region of interest are directly implemented in a single action thanks to the segmentation, but it remains entirely possible to initially detect the biometric trait, then in a second step to determine the region of interest associated with this biometric trait and extract it.

The detection of the biometric trait consists in practice in the generation of one or more "annotations" which locate in the first image the zone which contains said biometric trait, for example a box surrounding it, called detection box, in English " box” (defined by its size and position). We understand that the detection box is quite coarse and most often includes pixels that are clearly part of the background.

In the preferred case where the detection is more particularly a "segmentation", the final annotation generated is typically a "segmentation mask" (or segmentation map) of the first image, that is to say a binary image of same size as the original image. Specifically, pixels in the first image with a matching "white" pixel in the segmentation mask are considered biometric feature/ROI pixels, and pixels in the first image with a matching "black" pixel in the segmentation mask are considered background pixels.

Then, applying the mask to the first image isolates the biometric biometric feature/region of interest represented therein, in the same way that a user would circle it by hand, and so extract.

It is assumed that the user places his face in an appropriate manner in front of the first acquisition means 14a (in particular thanks to the guidance offered by the video feedback) so that there is only one detectable biometric feature and/or that the detection algorithm is able to choose a single "candidate" biometric trait like that of the user of the terminal 1, even if other faces would be visible in the background (these would therefore be considered as part of of the background).

Many detection/segmentation techniques suitable for separating the biometric feature/region of interest from the background are known to those skilled in the art. For example, suitable convolutional neural networks are known, such as U-Net, DeepLab, mask RCNN, YOLO, Inception, etc. who can be trained for this task. Reference may also be made to the document Li, H., Ngan, K. N., & Liu, Q. (2008). FaceSeg: automatic face segmentation for real-time video. IEEE Transactions on Multimedia, 11(1), 77-88.

Alternatively to semantic segmentation, it is possible to obtain and use a depth image also representing said individual representing said biometric trait, for example if the first acquisition means 14a have at least two juxtaposed cameras and consist of a "3D camera" thus making it possible to make stereovision.

By "depth image" or "depth map", we mean an image whose pixel value is the distance along the optical axis between the optical center of the sensor and the observed point. A depth image is sometimes represented (in order to be visually comprehensible) as a grayscale or color image whose luminance of each point is a function of the distance value (the closer a point is, the brighter it is ) but it will be understood that this is an artificial image as opposed to the first image which is a radiation image as defined above.

It will be understood that many other sensor technologies making it possible to obtain a depth image are known ("time of flight", sonar, structured light, etc.), and that in most cases, the depth image is in practice reconstructed by the processing means 11 from raw data provided by the first optical acquisition means 14a and which must be processed (we repeat that a depth image is an artificial object that a sensor cannot not easily obtained by direct measurement).

It is much easier to detect the biometric feature/ROI in a depth image than in the first image because:

• this is only slightly affected by exposure (the depth image does not include any information depending on the brightness);
• is very discriminating because it makes it possible to easily separate distinct objects, especially those in the foreground from those in the background (i.e. the background).

For this, said biometric feature/the region of interest is advantageously identified as the set of pixels of said depth image associated with a depth value included in a predetermined range, advantageously the closest pixels. This is a simple thresholding of the depth image, making it possible to filter out objects at the desired distance from terminal 1, possibly coupled with an algorithm making it possible to aggregate pixels into objects or blobs (to avoid having several “fragments” of the same biometric trait).

Preferably, the range [0; 1 m] or even [0; 50 cm] in the case of a wall-mounted terminal 1, but depending on the application, this range can be varied. It is then possible to extract the corresponding pixels from the first image.

In such an embodiment, the detection of the biometric feature can be implemented independently of the extraction of the region of interest; it is even preferred to implement a posteriori a detection algorithm such as those presented before (for example to CNN) on the identified region of interest.

Finally, whatever the technique for extracting the biometric feature, the second image is generated, the latter corresponding as explained to the first image in which the background has been at least partially replaced by said luminous background. For this, we “insert” for example the region of interest extracted on the luminous background.

Preferably, the second image corresponds to the first image in which everything except the region of interest has been replaced by said luminous background.

In the case where the region of interest coincides with the biometric line, the second image corresponds to the first image in which the background has been entirely replaced by said luminous background.

Remember that the brightness/color of the luminous background as well as the zoom level can be controlled to adjust the illumination provided by the luminous return.

It should be noted that having a region of interest "close" to the biometric trait, or even coinciding completely with the biometric trait, is of additional interest, namely limiting the information that can be used by a fraudster on the video feedback.

For example, a fraudster using a mask to try to fool the algorithm will not see the edge of his mask due to segmentation, and will not be able to estimate the quality of his fraud and attempt to improve it. On the other hand, the process always works on the first image (and not the second image) so that it can always implement effective anti-spoofing techniques.

As such, step (b) preferably further comprises verifying the authenticity of the detected biometric trait, using known techniques.

Indeed, spoofing mainly concerns flat or semi-flat masks which can deceive conventional detection algorithms, but which leave a suspicious "edge" of the mask visible.

Moreover, in a conventional manner, step (b) preferably comprises the prior adaptation of the exposure of the first image (before it serves as the basis for the second image) with respect to the biometric trait detected. To do this, as explained, the exposure of the entire image is normalized in relation to that of the zone considered: thus, we are sure that the pixels of the biometric feature are optimally exposed, if necessary to the detriment of the rest of the first image (i.e. the background), but this is irrelevant since we will overwrite it when generating the second image.

This is all the more effective if steps (a) to (c) are implemented cyclically since the exposure is thus progressively dynamically optimized at the same time as the first image is improved.

Authentication/identification

Finally, in a step (d), the authentication or identification strictly speaking of said individual is advantageously implemented on the basis of the biometric trait detected (in particular on the first image, if necessary improved by the illumination of the second image, as explained, it is understood that the second image is advantageously only used for video feedback from step (c) to improve the first image and not as a basis for biometric processing. In addition, step (b) comprises, as explained preferentially already the detection of the biometric trait of the individual in the first image so that said trait is directly available Alternatively, it remains however possible to make a new acquisition and/or a new independent detection).

More precisely, said biometric trait detected is considered as candidate biometric data, and it is compared with one or more reference biometric data from the database of the data storage means 12.

It then suffices to verify that this candidate biometric datum coincides with the/a reference biometric datum. In a known manner, the candidate biometric datum and the reference biometric datum coincide if their distance according to a given comparison function is less than a predetermined threshold.

Thus, the implementation of the comparison typically includes the calculation of a distance between the data, the definition of which varies according to the nature of the biometric data considered. The calculation of the distance includes the calculation of a polynomial between the components of the biometric data, and advantageously the calculation of a scalar product.

For example, in the case where the biometric data were obtained from iris images, a distance conventionally used to compare two data is the Hamming distance. In the case where the biometric data has been obtained from images of the face of individuals, it is common to use the Euclidean distance.

This type of comparison is known to those skilled in the art and will not be described in more detail above.

The individual is authenticated/identified if the comparison reveals a similarity rate between the candidate datum and the reference datum exceeding a certain threshold, the definition of which depends on the calculated distance.

Terminal

According to a second aspect, the present invention relates to the terminal 1 for implementing the method according to the first aspect.

The terminal 1 comprises data processing means 11, of the processor type, advantageously first optical acquisition means 14a (for the acquisition of the first), an interface 13, and if necessary data storage means 12 storing a reference biometric database.

The data processing means 11 are configured to implement:

• Obtaining, from the first optical acquisition means 14a of the terminal 1, a first image of said individual representing said biometric trait on a background;
• The generation of a second image, representing said biometric feature of the first image on a luminous background;
• the display of said second image on the interface 13 of said terminal 1 so as to illuminate the individual with said luminous background.

According to a third and a fourth aspect, the invention relates to a computer program product comprising code instructions for the execution (in particular on the data processing means 11 of the terminal 1) of a method according to the first aspect of the invention for acquiring a biometric trait of an individual, as well as storage means readable by computer equipment (a memory 12 of the terminal 2) on which this computer program product is found.

Claims (15)

Method for acquiring a biometric trait of an individual, characterized in that it comprises the implementation by data processing means (11) of a terminal (1) of the following steps:

1. Obtaining, from first optical acquisition means (14a) of the terminal (1), a first image of said individual representing said biometric feature on a background;
2. Generation, of a second image, representing said biometric feature of the first image on a luminous background;
3. Display of said second image on an interface (13) of said terminal (1) so as to illuminate the individual with said luminous background.

Method according to claim 1, comprising a new occurrence of step (a) while the second image is displayed on the interface (13). Method according to one of Claims 1 and 2, in which the second image corresponds to the first image in which the background has been at least partially replaced by the said luminous background, in particular entirely replaced by the said luminous background. Method according to one of Claims 1 to 3, in which step (b) comprises the prior extraction of a region of interest from the first image containing the biometric feature, the second image being generated from the region of interest extracted. A method according to claim 4, wherein said extracting a region of interest from the first image containing the biometric trait comprises detecting the individual's biometric trait in the first image, in particular by segmenting the first image. The method of claim 5, wherein step (b) further comprises adapting the exposure of the first image to the detected biometric feature prior to extracting the region of interest. Method according to one of Claims 5 and 6, further comprising a step (d) of authentication or identification of said individual on the basis of the biometric trait detected. A method according to claim 7, wherein step (d) comprises comparing the detected biometric trait with reference biometric data stored on data storage means (12). Method according to one of Claims 7 and 8, in which step (d) comprises the implementation of access control according to the result of said biometric identification or authentication. Method according to one of Claims 1 to 9, in which the said luminous background is a light plain background, preferably white. Method according to one of Claims 1 to 10, in which the said biometric feature of the individual is chosen from among a face and an iris of the individual. Method according to one of Claims 1 to 11, in which the said first image is a visible image. Terminal (1) comprising data processing means (11) configured to implement:

• Obtaining, from first optical acquisition means (14a) of the terminal (1), a first image of said individual representing said biometric feature on a background;
• The generation of a second image, representing said biometric feature of the first image on a luminous background;
• The display of said second image on an interface (13) of said terminal (1) so as to illuminate the individual with said luminous background.

Computer program product comprising code instructions for the execution of a method according to one of Claims 1 to 12 for acquiring a biometric trait of an individual, when said method is executed on a computer. Storage medium readable by computer equipment on which a computer program product comprises code instructions for the execution of a method according to one of Claims 1 to 12 for acquiring a biometric trait of an individual .