• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A biometric recognition method, comprising the step of calculating a similarity score of a candidate biometric vector with a reference biometric vector, one of the two biometric vectors being quantified.
    公开号:FR3083894A1
    申请号:FR1856529
    申请日:2018-07-13
    公开日:2020-01-17
    发明作者:Jonathan MILGRAM；Stephane Gentric
    申请人:Idemia Identity and Security France SAS；
    IPC主号:
    专利说明:

    The present invention relates to the field of biometric recognition, for example for the purposes of identifying an individual or verifying the individual's right to access a place or information.
    Technological background
    Conventionally, a biometric recognition method comprises the steps of:
    - capture an image of a body part of a recognition candidate;
    - extract from this image biometric characteristics in the form of a biometric vector of the candidate;
    - calculate a similarity score of the candidate's biometric vector with a reference biometric vector stored in a database hosted by a computer or in a data medium held by the candidate, such as an integrated circuit incorporated into a data card or an identity document such as a passport;
    - validate or refuse recognition based on the similarity score.
    The biometric reference vector is obtained during an enrollment operation which includes the steps of:
    - capture an image of a body part of an enrollment candidate;
    - extract from this image biometric characteristics in the form of a reference biometric vector stored in the database hosted by a computer or in the data medium held by the candidate.
    Currently, the reference biometric vector occupies a relatively large place in the memory of the data medium while the size of this memory is relatively limited.
    Subject of the invention
    An object of the invention is to provide a means to remedy the aforementioned problem or, more generally, at least partially limit the resources necessary for biometric recognition.
    Brief description of the invention
    To this end, a biometric recognition method is provided according to the invention, comprising:
    - during an enrollment phase, the steps of:
    - capture an image of an individual's body part;
    - extract from this image biometric characteristics in the form of a reference biometric vector;
    - during a recognition phase, the steps of:
    - capture an image of a body part of a recognition candidate;
    - extract from this image biometric characteristics in the form of a candidate biometric vector;
    - calculate a similarity score of the candidate biometric vector with the reference biometric vector, one of the two biometric vectors being quantified;
    - validate or refuse recognition based on the similarity score.
    Other characteristics and advantages of the invention will emerge on reading the following description of a particular non-limiting embodiment of the invention.
    Brief description of the drawings
    Reference will be made to the appended single figure schematically showing in perspective a device for implementing the invention.
    Detailed description of the invention
    The invention is here described in application to the access control to an airplane in an airport. Access to the plane requires having a boarding pass on which at least:
    - aircraft and flight identification elements;
    - the surname and first names of the passenger.
    The boarding card is traditionally issued by an automatic terminal from a reservation number or any other identifying information, or at the counter of the airline which chartered or owns the aircraft.
    In the mode of implementation which will be described, a reference biometric vector is associated with a personal document and more particularly here, with the boarding pass.
    This is perfectly achievable with a “paper” boarding pass but, in the implementation described here, the boarding pass is dematerialized.
    To this end, and with reference to the figure, the passenger having a mobile telephone 1 of the “ordiphone” or “smartphone” type is offered the download of an application making it possible to edit his boarding pass. This application is downloaded from a computer server 2 belonging to the airline which owns or has chartered the aircraft. The computer server 2 conventionally comprises a processor and a memory containing programs, and is connected to a data transport network 3, such as the Internet, to which the mobile telephone 1 can connect.
    The mobile telephone 1 is of the conventional type and comprises a processor, a memory containing programs (some of which are commonly called applications), and at least one image sensor 8.
    The computer server 2 executes a program comprising instructions for the sale of airplane tickets. To buy a plane ticket, the future passenger must identify himself on the computer server 2, select the desired flight, then pay. The program then offers to download the application to his mobile phone 1 so that he can create his boarding pass.
    When the downloaded application is executed, it commands the connection of the mobile phone 1 to the computer server 2 to:
    - execute a passenger authentication process (entering a username and password) then,
    - if the authentication is successful, execute a process for creating the boarding pass.
    The process of creation of the menu boarding includes the steps from: - capture a picture a part of body of a candidate to 1 ' enrollment r - extractof this picture of the
    biometric characteristics in the form of a so-called initial reference biometric vector;
    - quantifying the initial reference biometric vector to obtain a so-called quantified reference biometric vector;
    - form from the quantified reference biometric vector and an identifier of the flight a two-dimensional barcode.
    To capture the image, the application controls the image sensor 8 of the mobile phone 1 and asks the future passenger to take a photograph of his face (commonly called "selfie"). For this purpose, the application can display text or graphic indications allowing the future passenger to take a photograph of sufficient quality (in terms of exposure, sharpness, image contrast, face dimensions on the image. ..) to allow the extraction of biometric characteristics.
    The application then transfers, via the network 3, the captured image to the computer server 2, the program of which is arranged to detect the position of the face in the image, align the face in the image, then to perform the extraction of biometric characteristics. The detection of the position and the alignment of the face make it possible to obtain an image of predetermined dimensions cropped on the face: these operations are known in themselves and will not be detailed here.
    The extraction is carried out by at least one neural network trained by deep learning. The neural network is here a convolutional network which is configured to provide a reference vector having a size of 512 bytes. A single neural network is used here, but it is possible to use several neural networks for the extraction.
    We then obtain the initial biometric reference vector (commonly called "biometric template") which is then quantified to form the quantified reference biometric vector. The quantification used here is binarization. Remember that binarization is an operation consisting in comparing the value of each component of the quantified reference biometric vector to a threshold, and replacing the value of each component by 0 when the value of this component is less than or equal to the threshold and by 1 when the value of this component is greater than the threshold. For example, if the threshold is equal to 0, each component less than or equal to 0 is replaced by 0 and by each component greater than 0 by 1. Thus, if the components of the initial reference biometric vector are each coded out of four bytes, or 32 bits, and that each of its components is binarized, the quantized reference biometric vector resulting from the binarization of the initial reference biometric vector is 32 times smaller than that of the initial reference biometric vector. Alternatively, the threshold may be different for each component. Preferably then, the threshold corresponds to an average value of each component. These mean values were calculated, at the time of the design of the quantification algorithm, from a plurality of reference biometric vectors extracted from faces from a database of faces.
    At the end of the quantification, the quantified reference biometric vector has a size less than approximately 200 bytes and here, more particularly, a size less than 100 bytes or even less than 50 bytes.
    Other modes of quantification are possible. For example, one could divide the range of possible values of each value of the initial biometric reference vector into four sub-ranges and replace this real value (initially coded on 32 bits) by a quantized value equal to 0, 1, 2 or 3 according to the subinterval to which the actual value belongs. The quantized value is coded on only two bits.
    Preferably, the quantized reference biometric vector has a size of 32 bytes. The two-dimensional bar code (commonly called “QR code” and symbolized in Q in the figure) is formed in a conventional manner except that the information to be included therein is the quantified reference biometric vector and a signature (the signature has here a size of 16 bytes). The signature is an encrypted value corresponding here to the name of the passenger and to the flight information (in particular the flight identifier). Preferably, before the barcode is formed, a hash of the quantized reference biometric vector and of the signature is hashed. The barcode is then returned by the computer server 2 to the application.
    It will be noted that, after the capture, all or part of the operations allowing the constitution of the two-dimensional barcode can be carried out either within the mobile telephone 1 or the computer server 2, and in particular:
    - the extraction of the biometric characteristics can be carried out by the application within the mobile telephone 1 or by the computer server 2 after transmission of the captured image;
    - the same for the formation of the barcode.
    At the airport, the passenger is supposed to launch the execution of the application so as to allow the display of the virtual boarding pass. The application is also arranged to present, on request of the passenger, the barcode but also other information concerning the flight such as in particular the departure time. The application is also preferably arranged to connect to the computer server 2 of the airline or to a dedicated computer server 4 of the airport to which the computer server 2 is connected and to which the computer server 2 has communicated the information. relating to the passenger. This makes it possible to display on the mobile telephone 1, by way of notification or alert, the information relating to the check-in, the boarding gate, any delays ...
    The airport computer server 4 is connected to a terminal 5 for controlling access to the boarding area. Terminal 5 includes a processor and a memory for executing a control method, as well as a two-dimensional barcode reader 6 and at least one camera 7 arranged to capture images of the faces of passengers appearing before terminal 5 Terminal 5 is designed to command the opening of an access airlock under certain conditions.
    To access the boarding area, the passenger controls the application so that the bar code is displayed on the screen 9 of their mobile phone 1, which they present in front of the reader 6. Terminal 5 detects the bar code and controls the camera 7 to capture an image of the passenger's face.
    The terminal 5 control program then performs the steps of:
    - extract from this image biometric characteristics in the form of a candidate biometric vector;
    - read the barcode and extract the quantified reference biometric vector and the signature;
    - calculate a similarity score of the candidate biometric vector with the quantified reference biometric vector and decipher the signature;
    - validate or refuse recognition based on the similarity score, the accuracy of the passenger's name and the existence of an upcoming flight corresponding to the flight identifier;
    - open the access airlock if recognition is validated and issue an alert if recognition is refused.
    The calculation of the similarity score is a distance between the two vectors. More specifically, the calculation of the similarity score includes a comparison of the distance to a threshold defined as a function of a desired rate of false acceptance and a desired rate of false rejection.
    According to a first approach, the candidate biometric vector is binarized before the calculation of the similarity score. The similarity score here is a Hamming distance, the calculation of which is known in itself.
    The Hamming distance calculation is very fast. However, the binarization of the candidate biometric vector degrades the accuracy of recognition.
    According to a second approach, the candidate biometric vector, which contains only real values, remains unchanged (without quantification), which makes it possible to maintain good recognition precision. In a first version, the value of each component of the quantized reference biometric vector which is equal to 0 is replaced by -1 to obtain a transformed quantized reference biometric vector. The cosine distance separating the transformed quantized reference biometric vector and the candidate biometric vector is then calculated.
    The similarity score is then a cosine distance, the calculation of which is conventional.
    In a second version, during the design of the recognition algorithm, the average value μ ± and the standard deviation o ± of each component of biometric vectors extracted from a sample of images from a bank are calculated. of face data. Then, for each component of the quantified reference biometric vector:
    - if the value of component i is 0 so the value of the component is replaced through X ± = Pi - σ ± r- if the value of component i is 1 so the value of the component is replaced through
    x ± = μι + Oi.
    The similarity score is then a cosine distance calculated in a conventional manner.
    Of course, the invention is not limited to the embodiment described but encompasses any variant coming within the scope of the invention as defined by the claims.
    In particular, the reference biometric vector can be stored in graphic form (as in the two-dimensional barcode) but also in electronic form, for example in a memory of an integrated circuit of the RFID type contained in a document such as a passport, in a memory of an integrated circuit of an integrated circuit card, in a computer memory ...
    Although in the described implementation mode, it is the biometric reference vector which is provided by the neural network, it can be the candidate biometric vector.
    The extraction can be carried out in the mobile phone. This prevents biometric data from passing through a remote computer.
    Depending on the type or number of neural networks used, the size of the biometric vector may be different from that indicated. For example with two or three neural networks, the biometric vector can have a size of 1024 or 1536 bytes. An example of a neural network and a usable learning method is that described in the document "DeepVisage: Making face recognition simple yet with powerful generalization skills", Abul Hasnat et al., The IEEE International Conference on Computer Vision (ICCV), 2017, pp. 1682-1691.
    The two-dimensional barcode can be stored in the memory of an electronic device such as a telephone or multimedia electronic tablet, or printed on a document such as a “paper” boarding pass.
    It is possible to associate with the biometric reference vector mixed by hashing with the flight identifier also signature information which will include
    example a value integrity calculated sure the bits of vector biometric reference got after hash. Other information can to be mixed at vector biometric reference like through example a dated, the name of passenger or all other suite of
    alphanumeric character.
    The invention is applicable to other biometrics than face biometrics and for example fingerprints, irises ...
    Instead of being carried out from a mobile phone, enrollment can be carried out from a registration terminal provided with a biometric sensor and located for example at the airport.
    权利要求:
    Claims (15)
    [1" id="c-fr-0001]
    claims
    1. Method of biometric recognition, comprising : - during a phase c i 'enlistment, Steps from: - capture a picture of a part of body a individual; - extract of this picture of the biometric features under the form a biometric vector of reference;- during a phase of recognition, the steps from: - capture a picture of a part of body a candidate for recognition; - extract of this picture of the biometric features under the form a
    candidate biometric vector;
    - calculate a similarity score of the candidate biometric vector with the reference biometric vector, one of the two biometric vectors being quantified;
    - validate or refuse recognition based on the similarity score.
    [2" id="c-fr-0002]
    2. Method according to claim 1, in which the quantified biometric vector is the reference biometric vector.
    [3" id="c-fr-0003]
    3. Method according to claim 2, comprising the step, following the extraction of the reference biometric vector, of quantifying the reference biometric vector.
    [4" id="c-fr-0004]
    4. Method according to any one of the preceding claims, in which the calculation of the similarity score is a distance between the two vectors.
    [5" id="c-fr-0005]
    5. Method according to claim 4, in which the calculation of the similarity score comprises a comparison of the distance to a threshold defined as a function of a desired rate of false acceptance and of a desired rate of false rejection.
    [6" id="c-fr-0006]
    6. The method of claim 4, wherein the distance is a cosine distance.
    [7" id="c-fr-0007]
    7. Method according to any one of the preceding claims, in which the quantization applied to the biometric vector is a binarization.
    [8" id="c-fr-0008]
    8. The method of claim 7, comprising, before calculating the score, the step of replacing by -1 the value of each component of the quantified biometric vector which is equal to 0 to obtain a transformed quantized reference biometric vector.
    [9" id="c-fr-0009]
    9. The method as claimed in claim 7, comprising, before calculating the score, the step of replacing the value of each component of the quantified reference biometric vector:
    - by Xi = μ ± - σ ± if the value of the component i is 0; - by Xi = μ ± + σ ± if the value of the
    component i is 1;
    Pi and Oi being the mean value and the standard deviation of each component of biometric vectors extracted from a sample of face images.
    [10" id="c-fr-0010]
    10. Method according to any one of the preceding claims, in which the extraction is carried out by a neural network driven by deep learning.
    [11" id="c-fr-0011]
    11. Method according to the preceding claim, in which the neural network is a convolutional network.
    [12" id="c-fr-0012]
    12. The method of claim 10 or claim 11, wherein the neural network is configured to provide a reference vector having a size at most equal to approximately 512 bytes.
    [13" id="c-fr-0013]
    13. The method of claim 11, wherein the quantization is performed to provide a biometric vector having a size less than 200 bytes.
    [14" id="c-fr-0014]
    14. The method of claim 13, wherein the quantization is performed to provide a biometric vector having a size less than 100 bytes.
    [15" id="c-fr-0015]
    15. The method of claim 14, wherein the quantization is performed to provide a biometric vector having a size less than 50 bytes.
    类似技术:
    公开号 | 公开日 | 专利标题
    US9330142B2|2016-05-03|Biometric matching engine
    US9864982B2|2018-01-09|Image recognition-based payment requests
    US20170262472A1|2017-09-14|Systems and methods for recognition of faces e.g. from mobile-device-generated images of faces
    EP3594850A1|2020-01-15|A biometic recognition process
    EP3008704B1|2019-11-20|Method of control of persons and application to the inspection of persons
    US9069997B2|2015-06-30|Biometric matching system using input biometric sample
    EP2656321B1|2015-01-07|Process for enabling an authentication or an identification and corresponding verification system.
    US20210201061A1|2021-07-01|Intelligent gallery management for biometrics
    EP3249569B1|2018-05-23|Method for monitoring individuals with simplified authentication
    CN111886842A|2020-11-03|Remote user authentication using threshold-based matching
    CN111695099A|2020-09-22|Security system and corresponding access method, machine readable storage device
    KR20190122206A|2019-10-29|Identification methods and devices, electronic devices, computer programs and storage media
    EP3008703B1|2021-03-31|Biometric identification method
    EP3264329B1|2019-09-11|A method and a device for detecting fraud by examination using two different focal lengths during automatic face recognition
    JP2008009617A|2008-01-17|System, program, and method for individual biological information collation
    EP3594851A1|2020-01-15|Biometric recognition method
    CN111683285B|2021-01-26|File content identification method and device, computer equipment and storage medium
    US20220050890A1|2022-02-17|Remotely verifying an identity of a person
    CN111932363A|2020-11-13|Identification and verification method, device, equipment and system for authorization book
    Quinn et al.2013|Standard iris storage formats
    Timmerman et al.2020|Video Camera Identification from Sensor Pattern Noise with a Constrained ConvNet
    FR3100070A1|2021-02-26|Biometric recognition method with drift control and associated installation
    KR102047936B1|2019-11-22|Apparatus and method for classifying images stored in a digital device
    EP3467694A1|2019-04-10|Method and device for combined biometric recognition
    FR3105480A3|2021-06-25|Identification system
    同族专利:
    公开号 | 公开日
    FR3083894B1|2020-06-26|
    US20200019691A1|2020-01-16|
    US11074330B2|2021-07-27|
    EP3594850A1|2020-01-15|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US20030095689A1|2001-11-07|2003-05-22|Vollkommer Richard M.|System and method for mobile biometric authentication|
    JP6850817B2|2016-06-03|2021-03-31|マジック リープ， インコーポレイテッドＭａｇｉｃ Ｌｅａｐ，Ｉｎｃ．|Augmented reality identification verification|
    US10482336B2|2016-10-07|2019-11-19|Noblis, Inc.|Face recognition and image search system using sparse feature vectors, compact binary vectors, and sub-linear search|
    WO2019083508A1|2017-10-24|2019-05-02|Hewlett-Packard Development Company, L.P.|Facial recognitions based on contextual information|
    US10943096B2|2017-12-31|2021-03-09|Altumview Systems Inc.|High-quality training data preparation for high-performance face recognition systems|
    US20190311261A1|2018-04-10|2019-10-10|Assured Information Security, Inc.|Behavioral biometric feature extraction and verification|CN111428594A|2020-03-13|2020-07-17|北京三快在线科技有限公司|Identity authentication method and device, electronic equipment and storage medium|
    CN112037369A|2020-07-23|2020-12-04|汇纳科技股份有限公司|Unlocking method, system, medium and device of automatic parking spot lock based on vehicle identification|
    法律状态:
    2019-10-23| PLFP| Fee payment|Year of fee payment: 2 |
    2020-01-17| PLSC| Publication of the preliminary search report|Effective date: 20200117 |
    2020-06-23| PLFP| Fee payment|Year of fee payment: 3 |
    2021-06-23| PLFP| Fee payment|Year of fee payment: 4 |
    优先权:
    申请号 | 申请日 | 专利标题
    FR1856529A|FR3083894B1|2018-07-13|2018-07-13|BIOMETRIC RECOGNITION PROCESS|
    FR1856529|2018-07-13|FR1856529A| FR3083894B1|2018-07-13|2018-07-13|BIOMETRIC RECOGNITION PROCESS|
    EP19185872.9A| EP3594850A1|2018-07-13|2019-07-11|A biometic recognition process|
    US16/510,399| US11074330B2|2018-07-13|2019-07-12|Biometric recognition method|
    [返回顶部]