• 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:
    Method and apparatus for biometric corneal recognition. Corneal biometric recognition method comprising: (i) capturing at least one image of the corneal surfaces; and (ii) quantifying the topographic irregularities of said corneal surfaces to obtain characteristic and differential parameters of the cornea topography for at least one subject; and where each capture is configured as a record with at least one ocular parameter; and comprising a step of comparing: at least a first record with at least one ocular parameter; and at least a second record of the same ocular parameter; and where it also comprises a step of classifying the data resulting from the comparison by means of a classification algorithm with self-learning. (Machine-translation by Google Translate, not legally binding)
    公开号:ES2662912A1
    申请号:ES201800020
    申请日:2018-01-24
    公开日:2018-04-10
    发明作者:Celia Sánchez Ramos；Cristina BONNIN ARIAS；Javier PASCAU GONZÁLEZ-GARZÓN；Rafael MORETA MARTÍNEZ
    申请人:Universidad Complutense de Madrid；Universidad Carlos III de Madrid；
    IPC主号:
    专利说明:

    5
    10
    fifteen
    twenty
    25
    30
    35
    METHOD AND APPARATUS FOR CORNEAL BIOMETRIC RECOGNITION
    It is an object of the present invention a method and an apparatus for the biometric recognition of living beings (people or animals) that incorporates the topography of the cornea as a biometric constant characteristic of each individual.
    Prior art
    The types of recognition of people called biometrics are based on the physical characteristics of the user to be identified, although the authentication of users by means of biometric methods is possible using any unique and measurable characteristic of the individual, traditionally it has been based on six large groups: eye-iris , eye-retina, fingerprints, hand geometry, writing-signature and voice. However, biometric authentication models based on eye patterns are divided into different technologies: either they analyze retinal patterns, or they analyze the morphological characteristics of the iris.
    The iris is the most visible part of the human eye. It is a complex pattern that contains many distinctive aspects, such as wrinkles, arcuate ligaments, ridges or rings. The iris image can be taken from a reasonable distance (approximately 1 meter) with a high level of accuracy, although it is difficult to obtain if the person does not stay close to the camera or is not interested in authentication. In addition, medical processes can alter the color somewhat (not the texture), the humidity of the iris can cause reflection, the eyelids can partially cover the iris in certain people or human groups (for example, in people with slanted eyes) and can be manipulated by contact lenses, hypothetical trauma or suffer size changes due to pupil dilation.
    The retinal scanner is a biometric technique that uses retinal patterns to identify a person. Authentication of the retina can be done through the initial registration of the vascular structure of the retina, that is, the shape of the blood vessels of the human retina, which has characteristic elements of each individual and differential from the rest of the population. As an example of these methods we have documents US2015 / 0193666; US2015 / 0186721; US2015 / 0078630; US2014 / 0294252; US2014 / 0270405; US2014 / 0198959; US2014 / 0044321; US2014 / 0044320; US2014 / 0044319; US8483450; US8437513 and US8369595.
    5
    10
    fifteen
    twenty
    25
    30
    35
    In these systems, the user to be identified must look through an eye device, adjust the distance and movement of the head, look at the determined fixing point and, finally, press a button to indicate to the device that it is ready to the analysis. To establish valid records, you have to wait five minutes for the mydriasis, or pupil dilation necessary in the entrance systems by the pupil or use mydriatic drugs.
    Subsequently, the retina is scanned with a low intensity infrared radiation in the form of a spiral, detecting in an image the nodes and branches of the retina area to compare them with those stored in a database, if the sample matches the one stored by the user that the individual claims to be, authentication is validated. However, recognitions through the iris and the retina have similar limitations and new systems and methods that overcome the disadvantages of current methods remain necessary.
    EP2319392, in order to overcome the aforementioned drawbacks, provides a method and a biometric recognition system based on the analysis of the irregularities of the surface map of the second ocular diopter with respect to a normalized surface, which has as its main advantage the inability to access the second ocular diopter, neither by an external individual, nor by the person himself to recognize, since it is inside the eye. This document is based on taking an image exclusively of the second ocular diopter in order to determine the irregularities of its surface map with respect to a normalized surface and quantify them, resulting in a set of characteristic features of each person that can be used as an authentication system , that is, using the variations and calculations made with them as biometric minutiae. The image is proven safe as it is non-invasive, can be taken without any contraindication and as many times as necessary.
    Explanation of the invention.
    The object of the present invention is a method and an apparatus for the biometric recognition of people that incorporates, as a biometric constant, irregularities in the topography of the cornea, improving the technique described in EP2319392. Thus, the recognition of the person is carried out through the capture of a real-time image of the topography of the cornea, for which the device has a rotating camera that makes a three-dimensional image of the cornea. In a particular embodiment the camera is a camera of
    5
    10
    fifteen
    twenty
    25
    30
    35
    Scheimpflug. Subsequently, the first and second diopter maps of the ocular optical system are generated from the measurements made on the anterior and posterior surfaces of the cornea.
    The maps generated by the camera are corrected mathematically considering the optical effect of the first diopter in the second diopter. Finally, compared to EP2319392, where the surface maps of the second ocular diopter are compared with a standard surface, in the present invention a standardized record comparison system of ocular parameters is used (front and rear surfaces, ocular refractive power, or any other medium through the camera) to obtain a series of standardized characteristics of the comparison (mathematically, a numerical matrix) to subsequently establish a process of mathematical classification of the standardized characteristics of the comparison, so that the result of This classification process indicates who the person is, or if the person is who they say they are. The mathematical classification process also has the particularity of learning during its operation, so that, the greater the number of records, the more robust the classification process will be.
    Throughout the description and the claims the word "comprises" and its variants are not intended to exclude other technical characteristics, components or steps. For those skilled in the art, other objects, advantages and features of the invention will be derived partly from the description and partly from the practice of the invention. The following examples and drawings are provided by way of illustration, and are not intended to restrict the present invention. In addition, the present invention covers all possible combinations of particular and preferred embodiments indicated herein.
    Brief description of the drawings
    Next, a series of drawings that help to better understand the invention and that expressly relate to an embodiment of said invention that is presented as a non-limiting example thereof is described very briefly.
    FIG. 1 Shows the block diagram of example 1 of application of the method of the invention.
    FIG.2 Shows the block diagram of example 2 of application of the method of the invention.
    5
    10
    fifteen
    twenty
    25
    30
    FIG. 3 Shows the block diagram of example 3 of application of the method of the invention.
    Statement of a detailed embodiment of the invention
    EXAMPLE 1, Application of the recognition method for the identification of a subject
    As can be seen in FIG. 1, a record of the cornea of a subject is taken through a conventional Scheimpflug chamber and, subsequently, a plurality of ocular parameters are extracted from the surface maps of the cornea. In a particular embodiment, the ocular parameters employed are the anterior curvature, the posterior curvature of the cornea and the total refractive power of the cornea.
    Each measurement or capture lasts approximately 30 seconds and is constituted in a register, so that for each subject and ocular parameter "n" registers are established, since in this way the possible errors and false positives in the measurement are minimized .
    Subsequently, once the maps of each ocular parameter are established, the record A of a subject is compared with a second register B of the same ocular parameter to check whether it corresponds to the same subject or to a different subject. This comparison results in a numerical matrix. To make the comparison, parameter by parameter, different mathematical methods can be used, at least one of the following:
    Correlation coefficient:
    Two registers (A, B) of the same ocular parameter intersect.
    Only the values of the intersection area are chosen.
    A numerical value is calculated by applying the following formula:
    The following functions are calculated for those points (subscript n) of intersection between the A and B records.
    An = value of register A at point n Bn = value of register B at point n
    N = total number of intersection elements between the records A and B
    r =
    £ „(/!„ - A) (Bn - B)
    (E „G4„ - A) 2) (E „(fi„ "S) 2)
    Sum of absolute differences:
    Two registers (A, B) of the same ocular parameter intersect. Only the values of the intersection area are chosen.
    A numerical value is calculated by applying the following formula:
    AD =
    image 1
    image2
    Sum of squared distances:
    10 Two registers (A, B) of the same ocular parameter intersect.
    Only the values of the intersection area are chosen. A numerical value is calculated by applying the following formula:
    SSD = - ^ (A - Bn) 2
    n
    fifteen
    twenty
    Histogram comparison (Jeffrey divergence):
    The histogram is calculated for each record.
    A numerical value is calculated by applying the following formula:
    Jeffry's Divergence is calculated from the histograms of each record. He
    index i represents the intensity values.
    hA¡ = histogram of A for the intensity value i
    hB¡ = histogram of B for the intensity value i
    D =
    image3
    + hBi log
    image4
    Comparison of level maps:
    A plurality of levels are established for each record.
    5
    10
    fifteen
    twenty
    25
    30
    Each level in register A is compared separately with its corresponding level in register B.
    A Jaccard index is established for comparatives:
    The Jaccard index is calculated from two binary images (X and Y) to give an overlap value. It is obtained by dividing the cardinality of the intersection of both by the cardinality of their union.
    x nY
    J | xuy |
    The Jaccard index is rounded to 0 and 1.
    The rounded coefficients are added.
    Rounded coefficients are divided by the number of levels.
    It should be noted that these methods are not exclusive to each other, but can be applied interchangeably, one by one, or in combination with each other. Once the comparison is established, it is necessary to classify said data mathematically to establish who the subject is (identification), or in a particular case, described in example 2, to establish whether the subject is who he claims to be (authentication).
    For this, it is possible to implement any one of the classification algorithms with the following self-learning:
    Nearest Neighbors ( https://es.wikipedia.Org/wiki/K nearest neighbors)
    SVM linear ( https://es.wikipedia.org/wiki/ Support vector machines)
    RBF SVM ( https://en.wikipedia.org/wiki/Radial basis function kernel)
    Decision tree ( https://es.wikipedia.org/wiki/Appendix based on decision trees) Random Forest ( https://es.wikipedia.org/wiki/Random forest)
    Neural Network ( https://es.wikipedia.org/wiki/Artificial Neural Network)
    AdaBoost ( https://en.wikipedia.org/wiki/AdaBoost)
    Naive Bayes ( https://es.wikipedia.org/wiki/ Naive Bavesian Classifier)
    QDA ( https://en.wikipedia.org/wiki/Quadratic classifier)
    In any of the above algorithms a learning phase is required that is performed
    5
    10
    fifteen
    twenty
    25
    30
    35
    from a group of records of a plurality of subject (training group) from which the decision rules of each classification algorithm are generated. However, it has been observed that the classification with artificial neural networks is the one that offers the best results.
    Neural networks are a computational model based on a large set of simple neuronal units. The neural units are grouped in layers and each one is connected with neural units of the next layer (if it exists). Each neuronal unit, individually, operates using sum functions based on the information it receives from the neural units connected to it. There may be a limiting function in each connection and in the unit itself, so that the signal must exceed a limit before propagating to another neuron, increasing or inhibiting the activation state of adjacent neurons. This limiting function or activation function allows complex processes to be modeled non-linearly. There are multiplicative weights in each connection link between neuronal units. These weights are defined iteratively based on the result you want to obtain with a self-learning process.
    An application of artificial neural networks comprises several phases or stages. The following phases are carried out for the development and validation of the structure:
    Definition of the neural network: the number of neurons and layers is determined, as well as the different activation functions.
    Training of the neural network: the type of training that will be carried out is defined, in turn the learning algorithms are determined. Learning is a process of adjusting the weights between the layer connections. This adjustment is made from training data, so that the weights are modified iteratively until the network classifies training data in the best possible way.
    Use of the neural network: once the training phase has been completed, the network responds to an input stimulus by applying the weights calculated in the training phase, resulting in a classification. In this use a weight adjustment is not made.
    Maintenance of the neural network: over time if new sets of
    5
    10
    fifteen
    twenty
    25
    30
    data, it would be necessary to validate the architecture to ensure good use, even make a new learning.
    The neural network allows modeling different classification problems without too many restrictions regarding their characteristics. This may be the reason why the best results of classification of cornea records are obtained with this method.
    EXAMPLE 2. Application of the recognition method for the authentication of a subject
    As can be seen in FIG. 2, it is possible to implement a second layer for checking the identity of a subject, i.e. to know if a subject is who he says he is and not another. Thus, the corneal parameters of a subject A are entered into the previously trained classifier algorithm (preferably a neural network) as described in the previous example, which compares them to all stored records of said subject A. Subsequently, the number of matching records between subject A and the stored records of said subject A, such that, if this number is equal to or greater than a pre-set threshold of matching records, it is determined whether subject A is stored subject A in a database connected to the apparatus of the invention.
    EXAMPLE 3. Convolutionary Neural Network (CNN)
    As it is possible to observe in FIG. 3 and as a possible alternative to example 1 to determine if two registers are of the same subject or not, use can be made of convolutional neural networks (CNNs). Once the records of a subject A and a Subject B are obtained, they are introduced, without making any prior comparison, into the algorithm based on convolutional neural networks. The result of this would be directly the classification of the two records as the same subject. This algorithm also needs a learning process (similar to conventional neural networks) performed with a group of records of a plurality of subjects (training group). However, the phase of obtaining characteristics from the comparison between the records during any phase of the algorithm would not be necessary, since during training the CNN calculates the convolutional filters that detect which characteristics of the images are suitable for classification. .
    5
    10
    fifteen
    twenty
    25
    30
    35
    1. A corneal biometric recognition method comprising: (i) capturing at least one image of the corneal surfaces; and (ii) quantify the topographic irregularities of said corneal surfaces to obtain characteristic and differential parameters of the corneal topography for at least one subject; and where each capture is configured as a record with at least one ocular parameter; and characterized in that it comprises a stage of comparison between:
    (a) at least one first record with at least one ocular parameter; Y
    (b) at least a second record of the same ocular parameter;
    and where it also includes a classification stage of the data resulting from the comparison using a classification algorithm with self-learning.
    2. The method according to claim 1, wherein the comparison of records is performed by at least one algorithm selected from: correlation coefficient, sum of absolute differences, sum of squared differences, histogram comparison and map comparison of level.
    3. The method according to claim 1 wherein the classification with self-learning is performed by at least one algorithm selected from: closest neighbors, linear SVM, RBF SVM, Decision Tree, Random Forest, Neural Network, AdaBoost, Naive Bayes or QDA.
    4. The method according to claim 3 wherein the classification is performed by a trained neural network.
    5. Method according to claim 1 wherein the stages of comparison and classification are performed by at least one deep learning algorithm that includes a convolutional layer integrated in a multilayer convolutional neural network.
    6. The method according to claim 1 wherein the number of similar records between a first subject and the stored records of said first subject is calculated, such that, if the number of matching records is equal to or greater than a predetermined threshold of matching records, it is determined whether the first subject is the subject stored in a database.
    7. A corneal biometric recognition apparatus comprising a camera comprising an electronic device with a processor, a memory and one or more programs, in which the program or programs are stored in the memory and 5 configured to be executed by the processor; and characterized because
    Programs include instructions for executing the method according to any of claims 1 to 6.
    8. Use of the apparatus of claim 7 for the authentication of a subject.
    10
    9. Use of the apparatus of claim 7 for the identification of a subject.
    image5
    Suj B Label SSD SAD | r | D
     Yes I Registered  YES Reg2 1 0.08 0.06 0.98 0.7
     Yes I Registered  S2 Regí 0 0.60 0.55 0.4 0.25
    FIG. 1
    Suj A
    image6
    FIG. 2
    image7
    权利要求:
    Claims (9)
    [1]
    5
    10
    fifteen
    twenty
    25
    30
    35
    1. A corneal biometric recognition method comprising: (i) capturing at least one image of the corneal surfaces: and (ii) quantifying the topographic irregularities of said corneal surfaces to obtain characteristic and differential parameters of the corneal topography for the less a subject; and where each capture is configured as a record with at least one ocular parameter; and characterized in that it comprises a stage of comparison between:
    (a) at least one first record with at least one ocular parameter; Y
    (b) at least a second record of the same ocular parameter;
    and where it also includes a classification stage of the data resulting from the comparison using a classification algorithm with self-learning.
    [2]
    2. The method according to claim 1, wherein the comparison of records is performed by at least one algorithm selected from: correlation coefficient, sum of absolute differences, sum of squared differences, histogram comparison and map comparison of level.
    [3]
    3. The method according to claim 1 wherein the classification with self-learning is performed by at least one algorithm selected from: closest neighbors, linear SVM, RBF SVM, Decision Tree, Random Forest, Neural Network, AdaBoost, Naive Bayes or QDA
    [4]
    4. The method according to claim 3 wherein the classification is performed by a trained neural network.
    [5]
    5. Method according to claim 1 wherein the stages of comparison and classification are performed by at least one deep learning algorithm that includes a convolutional layer integrated in a multilayer convolutional neural network.
    [6]
    6. The method according to claim 1 wherein the number of similar records between a first subject and the stored records of said first subject is calculated, such that, if the number of matching records is equal to or greater than a predetermined threshold of matching records, it is determined whether the first subject is the subject stored in a database.
    [7]
    7. A corneal biometric recognition apparatus comprising a camera comprising an electronic device with a processor, a memory and one or more programs, in which the program or programs are stored in the memory and 5 configured to be executed by the processor; and characterized because
    Programs include instructions for executing the method according to any of claims 1 to 6.
    [8]
    8. Use of the apparatus of claim 7 for the authentication of a subject.
    10
    [9]
    9. Use of the apparatus of claim 7 for the identification of a subject.
    类似技术:
    公开号 | 公开日 | 专利标题
    Goyal et al.2018|Dfunet: Convolutional neural networks for diabetic foot ulcer classification
    Czajka2015|Pupil dynamics for iris liveness detection
    ES2524267T3|2014-12-04|Biometric recognition by studying the surface map of the second ocular diopter
    US10441160B2|2019-10-15|Method and system for classifying optic nerve head
    KR101790719B1|2017-10-26|Method for identificating Person on the basis gait data
    Fatima et al.2013|A secure personal identification system based on human retina
    Homayon2015|Iris recognition for personal identification using LAMSTAR neural network
    ES2662912B2|2019-09-13|Method and apparatus for corneal biometric recognition
    Boyd et al.2020|Post-mortem iris recognition—a survey and assessment of the state of the art
    CN109886080A|2019-06-14|Human face in-vivo detection method, device, electronic equipment and readable storage medium storing program for executing
    Czajka2016|Iris liveness detection by modeling dynamic pupil features
    Jia et al.2018|Biometric recognition through eye movements using a recurrent neural network
    Trokielewicz et al.2017|Iris and periocular recognition in arabian race horses using deep convolutional neural networks
    Manjunath et al.2018|Analysis of unimodal and multimodal biometric system using iris and fingerprint
    Passarella et al.2013|Development of iridology system database for colon disorders identification using Image processing
    Bevilacqua et al.2017|A computer aided ophthalmic diagnosis system based on tomographic features
    Mohammadi et al.2014|Real-time evaluation of breast self-examination using computer vision
    Azimi et al.2019|The effects of gender factor and diabetes mellitus on the iris recognition system’s accuracy and reliability
    Narote et al.2006|An automated iris image localization in eye images used for personal identification
    Ortega Hortas2009|Automatic system for personal authentication using the retinal vessel tree as biometric pattern
    Hutchison2018|A Longitudinal Analysis on the Feasibility of Iris Recognition Performance for Infants 0-2 Years Old
    HUSSEIN2017|Biometric identification system based on hand geometry/El geometrisine dayalı biyometrik tanımlama sistemi
    Barra2016|Design of a Multi-biometric Platform, based on physical traits and physiological measures: Face, Iris, Ear, ECG and EEG
    Trokielewicz et al.2018|DCNN-based Human-Interpretable Post-mortem Iris Recognition
    Malathi et al.2020|IRIS PATTERN RECOGNITION: A QUANTITATIVE REVIEW ON HOW TO IMPROVE IRIS-BASED PATTERN RECOGNITION SECURITY SYSTEM
    同族专利:
    公开号 | 公开日
    ES2662912B2|2019-09-13|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    ES2224838A1|2003-02-21|2005-03-01|Universidad Politecnica De Madrid|System for biometric identification of people, by analyzing iris, for use in access control in e.g. buildings, has determination unit determining signature of iris, and comparison unit comparing captured image with stored image|
    US20110058712A1|2008-07-24|2011-03-10|Universidad Complutense De Madrid|Biometric recognition through examination of the surface map of the second ocular dioptric|
    WO2015102704A2|2013-10-08|2015-07-09|Sri International|Iris biometric recognition module and access control assembly|
    US20160012292A1|2013-10-08|2016-01-14|Sri International|Collecting and targeting marketing data and information based upon iris identification|
    WO2018013199A1|2016-07-14|2018-01-18|Magic Leap, Inc.|Iris boundary estimation using cornea curvature|
    法律状态:
    2019-09-13| FG2A| Definitive protection|Ref document number: 2662912 Country of ref document: ES Kind code of ref document: B2 Effective date: 20190913 |
    优先权:
    申请号 | 申请日 | 专利标题
    ES201800020A|ES2662912B2|2018-01-24|2018-01-24|Method and apparatus for corneal biometric recognition|ES201800020A| ES2662912B2|2018-01-24|2018-01-24|Method and apparatus for corneal biometric recognition|
    [返回顶部]