Method and system for classifying dactylograms

Abstract

Computer-implemented method (3000) for classifying dactylograms from a plurality of membership classes C_j, each of the classes corresponding to a particular anatomical area Z_i of the palmar surface of a hand, said method (2000) takes, as input data (I3000), at least one dactylogram D, and provides, as output data (O3000), a membership class C or a list L of membership classes C_k of the dactylogram D from the plurality of membership classes C_j.

Description

Title of the invention: Method and system for classifying fingerprints Technical field

[0001] The invention relates to a method and system for classifying fingerprints. The invention enables automatic identification of the type of fingerprints acquired by a fingerprint acquisition device. Technical background

[0002] Dactylograms, generally better known as "fingerprints" and / or "palmprints", are drawings formed by the traces left by dermatoglyphs of the fingers and / or palms of the hand on surfaces. Dermatoglyphs are the superficial furrows formed on the palms, soles and fingertips by dermal ridges and arranged in lines or spirals. They are unique to each individual and the drawings they form constitute a biometric "identity card" by which they can be identified.

[0003] Fingerprinting is a common practice in various administrative procedures with state institutions and in operations carried out by law enforcement agencies with a suspect or accused in the context of an offense, misdemeanor or crime. This survey is generally carried out using suitable fingerprint acquisition devices.

[0004] These acquisition devices generally consist of an electronic box comprising an acquisition surface equipped with a sensor on which one or more fingers, the hand or a part of the hand are placed to acquire an image of the dermatoglyphs.

[0005] US 2012 014569 Al, IB KOREA LTD [KR], 19.01.2012 describes a portable device for acquiring dactylograms. The device comprises an acquisition surface provided with an electroluminescent sensor on which the fingers of a hand can be placed to acquire an image of their dermatoglyphs.

[0006] US 2017 046554 Al, NEC CORP [JP], 16.02.2017 describes a portable device for acquiring digital fingerprints. This device is configured to provide instructions allowing a user to correctly acquire fingerprints.

[0007] Today, fingerprints are biometric information that is very widely used to identify individuals and / or authenticate a transaction such as an internet banking transaction or the use, by an individual, of a password among those stored in a password wallet. However, unlike authentication where a fingerprint acquired for an individual is compared to a alone or to a very limited number of reference fingerprints (1:1), the identification of an individual from a fingerprint requires the comparison of this fingerprint with numerous fingerprints previously acquired from several individuals (1:N) and generally stored in a database.

[0008] Because dactylograms are drawings with complex characteristics and the number of comparisons required during identification can become very high, the comparison of dermatoglyphs can be lengthy despite the computing resources of data processing devices available today. In order to reduce the time required to carry out these operations, it is known to classify dactylograms according to different classes based on certain morphological characteristics of dermatoglyphs. As examples, these morphological characteristics can be the general shape of the dermatoglyph (orientation of loops, arches, spirals, etc.) according to the categories of Henry Faulds, Francis Galton and Edward Henry, the overall outline of the ridges, the "minutiae" constituted by singular points along the ridges (termination of a ridge, bifurcation, etc.), the shape of the ridges, the pores or even the scars.

[0009] US 5572597 A1, LORAL CORP [US], 05.11.1996 describes a method for classifying digital fingerprints into different types based on the overall patterns formed by the dermatoglyphs, including ridges and furrows, of the entire fingerprints. The classification method is based on an analysis of the directions and angles formed by the local patterns of ridges and furrows in small areas of interest of the fingerprints.

[0010] EP 0 779 595 A2, NEC CORP [JP], 18.06.1997 describes a method for classifying digital fingerprints into five categories derived from the categories of Henry Faulds, Francis Galton and Edward Henry: simple arc, straight arc, right loop, left loop and spiral. The method is based on a combination of the results of two classifiers to calculate a probability of a fingerprint belonging to one of these four categories.

[0011] US 5,825,907 Al, LUCENT TECHNOLOGIES INC [US], 20.10.1998 describes a method for classifying finger prints into five categories derived from the categories of Henry Faulds, Francis Galton and Edward Henry: right loop, left loop, arc, spiral and double loop. The method uses an artificial neural network configured to classify finger prints based on a map of local groove shapes.

[0012] Wang et al., Fingerprint Classification Based on Depth Neural Network, arXiv preprint arXiv: 1409.5188, 2014 describes a method for classifying finger prints according to the four categories of Edward Henry: arc, right loop, left loop and spiral. The method implements, in a first step, an artificial neural network trained using unsupervised learning to extract the orientation map of a digital fingerprint - the orientation map of a digital fingerprint corresponds to the general direction pattern of the ridges of the dermatoglyph. In a second step, a supervised trained logistic regression model for "fuzzy" classification calculates the probability of the orientation map belonging to one of the four categories.

[0013] US 9,530,042 Bl, UNIV KING SAUD [SA], 27.12.2016 describes a method for classifying fingerprints into four categories derived from the categories of Henry Faulds, Francis Galton and Edward Henry: right loop, left loop, arc and spiral. The method first implements an operation of calculating a feature vector for each fingerprint using a descriptor based on a directional binary model with local gradient. The vectors are then processed by an artificial neural network configured to classify the fingerprints into one of the four categories.

[0014] Michelsanti et al., Fast Fingerprint Classification with Deep Neural Networks, VISIGRAPP 2017, describes a method for classifying fingerprints according to the four categories of Edward Henry: arc, right loop, left loop and spiral. The method implements a convolutional network of artificial neurons of the VGG-F or VGG-S type trained on a set of fingerprints classified according to the four categories. This method allows the direct extraction of fingerprint characteristics and therefore dispenses with an intermediate step for this extraction.

[0015] EP 3 825 915 Al, IDEMIA IDENTITY & SECURITY FRANCE [FR], 26.05.2021 describes a method for classifying digital fingerprints according to a given number of categories derived from the categories of Henry Faulds, Francis Galton and Edward Henry. The method implements a convolutional network of artificial neurons trained to determine whether a digital fingerprint belongs to each category. A fingerprint may possibly belong to several categories. Summary of the invention Technical problem

[0016] Depending on the contexts of operation and use - for example, a civil framework for administrative procedures and / or border crossings; a criminal framework for recording the fingerprints of a suspect or accused - the acquisition of types of fingerprints, in addition to digital fingerprints, may be relevant because they contain additional biometric information that can be used as more precise and supplementary means of identifying or authenticating an individual. Also, many devices, such as those described above, are they arranged to acquire, in addition to finger prints, other types of prints such as full and / or partial palm prints, or prints of several phalanges or several fingers.

[0017] Like digital fingerprints, for subsequent efficient exploitation of these other types of fingerprints, their classification is a prerequisite. Furthermore, because they are generally acquired on the same device during the same acquisition campaign, this classification must at least distinguish between these different types. However, the methods for classifying digital fingerprints are totally unsuitable for such an operation because they are exclusive of all fingerprints other than digital fingerprints.

[0018] A first negative consequence is that it is up to the operator to proceed himself to the distinction or classification of the different types of dactylograms during the acquisition campaign. However conscientious he may be, the risk of error remains both for the correct attribution of the dactylogram to a category and the identification of dactylograms not conforming to the type expected during acquisition.

[0019] A second negative consequence is the equally obvious inability of the aforementioned current methods to verify the integrity, namely the accuracy, completeness and reliability, of the fingerprints in a database comprising several fingerprint types, and, if necessary, to propose corrections. More precisely, these methods cannot verify that a partial palm fingerprint corresponding to the lower part of a palm has not been, through mistake or inadvertence, classified as a fingerprint of several fingers or phalanges.

[0020] There is therefore a need for a method for the reliable and automatic identification and / or classification of fingerprints of different types acquired during multimodal acquisition campaigns and / or using, in particular, multimodal devices such as described above. Technical solution

[0021] In a first aspect of the invention, there is provided a computer-implemented method for classifying dactylograms among a plurality of membership classes, each of which corresponds to a particular anatomical area of the palmar surface of a hand. Said method takes, as input data, at least one dactylogram, and provides, as output data, a membership class or a list of membership classes of the dactylogram among the plurality of membership classes. Said method comprises the following steps: - providing a convolutional network of artificial neurons, said convolutional network being previously trained on a training data set composed of a plurality of dactylograms classified according to a plurality of classes, each of which classes correspond to a particular anatomical area of a hand, said convolutional network being configured to provide the probabilities of membership of each dactylogram of the set E to each of the classes of the plurality of membership classes; - inferring, using the trained convolutional neural network, the probabilities of membership of the fingerprint provided as input data to each of the classes of the plurality of membership classes; - selecting a membership class from the membership classes for which the membership probability of the fingerprint provided as input data is the highest among the inferred membership probabilities for said classes, or a list of a number of membership classes selected from the plurality of membership classes and sorted according to an ascending or descending order of the membership probabilities to said inferred classes for the fingerprint provided as input data.

[0022] Advantageous embodiments of the first aspect of the invention are described in detail below.

[0023] In a second aspect of the invention, there is provided a data processing device comprising means for implementing a method according to the first aspect of the invention.

[0024] In a third aspect of the invention, there is provided a computer program comprising instructions which, when the program is executed by a computer, cause the computer to implement a method according to the first aspect of the invention.

[0025] In a fourth aspect of the invention, there is provided a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to implement a method according to the first aspect of the invention.

[0026] In a fifth aspect of the invention, there is provided a system for classifying among a plurality of membership classes, said system comprising a fingerprint acquisition device and a data processing device according to the second aspect of the invention. Brief description of the drawings

[0027] [Fig. 1] is an example of a fingerprint acquisition device.

[0028] [Fig.2] is a table grouping examples of schematic representations of classes of dactylograms in which each of the classes corresponds to a particular anatomical area of a hand.

[0029] [Fig.3] is a flowchart of a method according to the first aspect of the invention.

[0030] [Fig.4] is a schematic representation of an artificial neural convolutional network architecture according to an exemplary embodiment of a method according to the first aspect of the invention.

[0031] [Fig.5] is a representation of a data processing device according to the second aspect of the invention.

[0032] [Fig.6] is a representation of the classification performance of an example of a method according to the first aspect of the invention expressed as the number of classification successes and failures as a function of the probability of belonging to the class predicted by said method. Detailed description of the embodiments

[0033] [Fig. 1] schematically shows an example of a multimodal device 1000 for acquiring dactylograms making it possible to acquire, in addition to digital dactylograms, other types of dactylograms such as complete and / or partial palm dactylograms, or even dactylograms of several phalanges or several fingers. The device generally comprises an electronic box 1001 provided with an acquisition surface 1002 on which one or more fingers, the hand or a part of the hand can be placed to acquire an image of the dermatoglyphs. Many devices of this type are described in the state of the art.

[0034] The dactylograms acquired by a multimodal device such as that described above are generally in the form of an image of the drawings formed by the dermatoglyphs. These images may be single-channel images, for example, a grayscale image, or multi-channel images, for example RGB images.

[0035] Dactylograms can be classified according to several classes of membership, each of the classes corresponding to a particular atomic zone of the palmar surface of the hand to which they correspond. The term "palmar surface" of a hand is understood to mean the surface of the hand on which the palm of the hand is included, as opposed to the dorsal surface. The palmar surface includes the palm and all the fingers, in other words the thumb, index finger, middle finger, ring finger, little finger, thenar eminence, the hollow and the hypothenar eminence. This definition corresponds to that commonly accepted in anatomy.

[0036] The number of membership classes is not limited and may vary according to the needs and contexts of operation and use. Similarly, the nature of the membership classes, i.e. the anatomical areas of the palmar surface of the hand to which they correspond, may be defined or customized according to the same needs and contexts of operation and use.

[0037] Nevertheless, in practice, with reference to [Fig.2], it can be distinguished, among the different types of dactylograms, 11 Cl-Cll classes of membership corresponding to different anatomical areas of the palmar surface of the hand and adapted to a very large number of needs and use cases: the complete right hand (Cl), the complete left hand (C2), a lower area of the palm of the right hand (C3), a lower area of the palm of the left hand (C4), an upper area of the right hand (C5), an upper area of the left hand (C6), the right palm of the writer (C7), the left palm of the writer (C8), at least two, preferably three, preferably four fingers of the right hand (C9), at least two, preferably three, preferably four fingers of the left hand (CIO), the thumbs of the right hand and the left hand (Cil).

[0038] As examples, in a first recording framework for administrative procedures, it may be advantageous to acquire five different classes such as a lower area of the palm of the right hand, a lower area of the palm of the left hand, four fingers of the right hand, four fingers of the left hand, and both thumbs. On the other hand, in a second framework for recording the fingerprints of a suspect or an accused person, the number of types of fingerprints and therefore of classes of membership may be higher so as to obtain the most exhaustive biometric information possible. In addition to the previous classes, it includes one or more classes among the complete right hand, the complete left hand, an upper area of the right hand, an upper area of the left hand, the right palm of the writer, the left palm of the writer.

[0039] As explained above, for efficient exploitation and / or effective acquisition of the different types of fingerprints that may be acquired, it is necessary to classify them. However, current methods for classifying digital fingerprints are completely unsuitable, with the negative consequences described above.

[0040] Also, according to a first aspect of the invention, with reference to [Fig. 3], there is provided a computer-implemented method 3000 for classifying dactylograms among a plurality of membership classes Cj, each of the classes corresponding to a particular anatomical area Z_i of the palmar surface of a hand. The method 3000 takes, as input data 13000, at least one dactylogram D, and provides, as output data 03000, a membership class C or a list L of membership classes C_k of the dactylogram D among the plurality of membership classes Cj. The method 3000 comprises the following steps: - provide 3001 a CNN convolutional network of artificial neurons, said CNN convolutional network being previously trained on a set E of training data composed of a plurality of dactylograms D_i classified according to a plurality of classes Cj, each of the classes corresponding to an anatomical zone particular of a hand, said CNN convolutional network being configured to provide the membership probabilities Pj of each dactylogram D_i of the set E to each of the classes of the plurality of membership classes Cj; - infer 3002, using the trained CNN convolutional neural network, the membership probabilities P_i of the dactylogram D provided as input data 13000 to each of the classes of the plurality of membership classes Cj; - selecting 3003 a membership class C from the membership classes Cj for which the membership probability P of the dactylogram D provided as input data 13000 is the highest among the membership probabilities P_i inferred for said classes C_j, or a list L of a number k of membership classes C_k selected from the plurality of membership classes Cj and sorted according to an increasing or decreasing order of the membership probabilities P_k to said classes C_k inferred for the dactylogram D provided as input data 13000.

[0041] “Plurality of classes” means a set comprising at least two classes.

[0042] In practice, in the inference step 3002, at the output of the convolutional network CNN, a vector or a list V = {Pj} is provided, each of the values of which is the probability Pj of membership for each of the classes of the plurality of classes Cj of membership. There is thus a vector V for each dactylogram D provided as input data 13000.

[0043] In step 3001, according to a first alternative, a membership class C is selected from the membership classes Cj for which the membership probability P of the dactylogram D provided as input data 13000 is the highest among the membership probabilities P_i inferred for said classes C_j. This selection operation can be expressed according to the following mathematical formula: C -C^ (Cj, P; - max{P,}

[0044] According to a second alternative, a list L of a number k of membership classes C_k selected from the plurality of membership classes Cj and sorted according to a decreasing order of the probabilities P_k of membership to said classes C_k inferred for the dactylogram D provided as input data 13000 is selected.

[0045] A first remarkable advantage of the method according to the first aspect of the invention is to reduce the risk of error that an operator is likely to make when assigning a class to a fingerprint during or just after its acquisition. The method can thus be considered as providing a corrective, control or double-checking function during acquisition.

[0046] According to a first implementation example, the method according to the first aspect can provide a function for controlling the expected dactylogram class during a dactylogram acquisition campaign. By comparing the class that it assigns to a newly acquired dactylogram with the expected class for this dactylogram, it makes it possible to verify on the fly that the acquisition campaign is taking place in accordance with a previously established acquisition program such as the acquisition of a single given dactylogram class or with an order of acquisition of dactylograms of different classes. In the event of non-compliance with the expected, a warning signal can be sent to the operator by any appropriate means.For example, it may be expected, during an acquisition campaign, that several dactylograms of different classes are acquired in a certain order: four fingers of the right hand, then four fingers of the left hand, then the palm of the right hand, then finally the palm of the left hand. If, through an error in instructions on the part of the operator and / or through inadvertence on the part of the person for whom the dactylograms are acquired, a dactylogram of the palm of the left hand is acquired instead of the right, by comparing the class the method assigns to the dactylogram thus acquired with that expected, a warning signal may be sent to the operator so that he proceeds with a new acquisition.

[0047] According to a second implementation example, the method can provide an a posteriori corrective function. At the end of the acquisition of a fingerprint for which an operator must manually assign a membership class, the method can make it possible to verify that the class thus assigned by the operator is correct. For example, the method can verify that a fingerprint of a right palm has not been classified by the operator as a fingerprint of a left palm. In the event that the class initially assigned by the operator does not correspond to that which the method according to the invention would have assigned, a warning signal can be emitted by any appropriate means to the operator so that he can validate his classification or correct it on the basis of the class suggested by the method.As before, this double verification, here first human then machine, makes it possible to guarantee the accuracy of the classification of the fingerprints and to ensure the integrity of the databases in which the fingerprints are then likely to be stored.

[0048] According to a third implementation example, the method can provide an a priori corrective function. As soon as a fingerprint is acquired using a device, it can propose a membership class, or even several membership classes sorted according to a decreasing order of their membership probability value, to an operator who has the choice of validating or not this proposal, or of choose a class from those proposed. The risk of misclassification by the operator is thus reduced since he must make a choice for one class or a limited number of them. In addition, particularly in the event that the acquisition of a fingerprint takes place under difficult conditions, the risk of misclassification by the method itself can be eliminated by the intervention of the operator who is responsible for the final decision regarding the choice of the attribution class. This double verification, first machine then human, makes it possible to guarantee the accuracy of the classification of fingerprints and ensures the integrity of the databases in which the fingerprints are then likely to be stored.

[0049] A second remarkable advantage of the method according to the first aspect of the invention is to make it possible to verify the integrity of a database of fingerprints of different classes. For example, a database comprising fingerprints of different classes may contain classification errors. The method can be advantageously used to detect and correct these errors by providing it as input data each of the fingerprints in the database and by comparing the class that it attributes to it according to the greatest probability of belonging with that entered in the database for this fingerprint. In the event of a mismatch between the two classes, an automatic correction and / or the referencing of the fingerprint in an appropriate list can be carried out for subsequent checking by an operator.

[0050] According to certain preferred embodiments, with reference to [Fig. 3], the classes of the plurality of membership classes are selected from the complete right hand (C1), the complete left hand (C2), a lower area of the palm of the right hand (C3), a lower area of the palm of the left hand (C4), an upper area of the right hand (C5), an upper area of the left hand (C6), the writer's right palm (C7), the writer's left palm (C8), at least two, preferably three, preferably four fingers of the right hand (C9), at least two, preferably three, preferably four fingers of the left hand (C10), the thumbs of the right hand and the left hand (C11). The plurality of membership classes can thus comprise a number of classes equal to or greater than two up to all eleven membership classes.

[0051] The artificial neural convolutional network is of any type suitable for implementing the method according to the first aspect of the invention. According to certain preferred embodiments, the artificial neural convolutional CNN network comprises: - a first sequence S1 of convolutional layers of artificial neurons; - a second sequence S2 of residual convolutional blocks of artificial neurons; - a third sequence S3 comprising at least one layer of artificial neurons fully connected.

[0052] The first sequence SI of convolutional layers has the function, on the one hand, of reducing the dimensions of the dactylogram and, on the other hand, of increasing the number of topographic maps (or feature maps) relevant for the extraction of low-level patterns and features such as the distribution and intensity of colors, textures, shapes, and local contrasts. The second sequence S2 of convolutional blocks has the function of increasing the receptive field of the network while improving the extraction of low- and high-level features by combining the topographic maps of different layers through the residual connections.The last sequence S3 of fully connected artificial neuron layers has the function of combining all the features extracted by the two previous sequences to establish a probability of belonging of a dactylogram to each of the classes of the plurality of membership classes.

[0053] As explained previously, at the output of the sequence S3 of layers of artificial neurons, each of the values of the vector V = {P_j} of which is a probability Pj of membership for each of the classes of the plurality of classes Cj of membership. These values are specific to the convolutional network and are generally presented in the form of real numbers whose distribution differs from the usual probability distributions over an interval [-1; 1] or [0; 1]. Also, the vector V can be converted using a conversion function perhaps for example of the sigmoid type or, preferably of the “softmax” type.

[0054] In [Fig.4] is illustrated a detailed example of a convolutional CNN network suitable for the classification, among the 11 membership classes according to the embodiments described above, of fingerprints provided in the form of multi-channel images such as the RGB format with a resolution less than 1000x1000 pixels between 100 and 500 dpi. In detail, the CNN convolutional network comprises: - a first sequence SI of four successive convolutional layers whose dimension k of their kernel, the number n of instances of their kernel, the stride s and the padding p are respectively: (k = 5, n = 8, s = 2, p = 2), (k = 3, n = 16, s = 2, p = 1), (k = 3, n = 32, s = 2, p = 1) and (k = 3, n = 96, s = 2, p = 1), and each convolutional layer being followed by a ReLu type correction layer; - a second sequence S2 of three residual convolutional blocks of type “Depthwise Separable Convolutions”; each of the blocks comprising two sub-blocks SB1, SB2 with a residual connection at the output of the second sub-block, followed by a correction layer of type ReLu before the input of the next block; the first sub-block SB1 is composed of a convolutional layer with n = 96 instances of kernels of dimension k = 3, a step s = 1, and a padding p = 1, of a batch normalization layer ("batch normalization") and a ReLu type correction layer; the second sub-block SB2 is composed of a convolutional layer with n = 96 instances of kernels of dimension k = 3, a step s = 1, and a padding p=l, and a batch normalization layer ("batch normalization"); - a third sequence S3 comprising a layer of fully connected artificial neurons providing as output a vector of dimension 11, each of whose dimensions corresponds to each of the 11 membership classes described in the context of [Fig.2]; possibly, at the output of the second sequence S2 and before entering the third sequence, the topographic maps may be subject to a flattening operation.

[0055] The convolutional network CNN of artificial neurons represented in [Fig.4] being an example, it has no limiting character within the scope of the present invention. The number of layers in each of the sequences described previously, the dimensions of the convolution kernels, the number of their instances, the step and padding values can be adapted to the different needs and use cases of the fingerprints, in particular as regards the format, the resolution and the acquisition conditions of their images.

[0056] According to certain additional embodiments, the convolutional network CNN of artificial neurons further comprises, between the first sequence S1 and the second sequence S2, a spatial resampling step, preferably by bilinear interpolation. The function of this resampling step is to resize topological maps (or feature maps) obtained at the output of the second sequence into topological maps of fixed size before their entry into the third sequence, regardless of the size of the dactylograms provided as input to the method or the convolutional network of artificial neurons. The value of the fixed size to which the topographic maps are reduced corresponds to the number of input neurons of the third sequence comprising at least one layer of fully connected artificial neurons.

[0057] Thanks to this spatial resampling step, the method can take as input data dactylograms of any size. It can thus be advantageously implemented with any type of dactylogram acquisition device without it being necessary to carry out a resizing of the dactylograms that they are likely to provide. It can also be used for the classification of dactylograms contained in a database whose sizes are not homogeneous.

[0058] The training, also called learning, of the convolutional network of artificial neurons can be implemented in any suitable manner. According to certain advantageous embodiments, the cost function used during the training of the network The convolutional CNN of artificial neurons includes a cross-entropy cost function. Concretely, a cross-entropy function compares the distribution, q, of probabilities - in this case the probabilities P_j of belonging to each of the membership classes Cj - predicted by a model (the convolutional CNN of artificial neurons), with a distribution, p, of reference probabilities of a training data set on which the model is applied - in this case the probabilities, equal to 0 or 1, of belonging to the classes Cj of the dactylograms of a training set comprising previously classified dactylograms. The model is trained iteratively by minimizing the divergence between the two distributions. An example of a cross-entropy cost function, f, can be expressed as follows, with j the number of membership classes Cj: f = H ( p, q ) = - ^jP ( j ) log# ( j )

[0059] When, in accordance with certain embodiments described above, the convolutional network of artificial neurons comprises a sequence comprising at least one layer of fully connected artificial neurons, the cost function may further comprise a conversion function, real vectors outputting said layer into probability distribution over the interval [0; 1]. The conversion function may for example be of the sigmoid type or, preferably, of the “softmax” type. This function is applied before the cross-entropy function.

[0060] Depending on the needs and the use cases, the method according to the first aspect of the invention makes it possible to differentiate the dactylograms originating from the right hand from two from the left hand, and vice versa. It is preferably sensitive to chirality, that is to say to the axial symmetry relationship between one hand and the other, for example between a dactylogram of the right palm and that of the left palm. Thus, according to certain advantageous embodiments, the cost function used during the training of the CNN convolutional network of artificial neurons comprises a function of sensitivity to axial symmetry.

[0061] Such an axial symmetry sensitivity function can be implemented, during training, in the following manner. For each dactylogram of a training set, a symmetrical dactylogram is first created, obtained by an axial symmetry operation of said dactylogram. Each dactylogram and its symmetrical are then provided as training input data to the convolutional artificial neural network to calculate the probabilities of belonging to each of the classes of the plurality of membership classes. These membership probabilities are presented, at the output of the network, in the form of a vector whose values correspond to the probability of membership for each of the classes of the plurality of membership classes. There is thus a vector V[D_i] for each dactylogram D_i of the training set and a vector Vs[sD_i] for each corresponding symmetric dactylogram sD_i.

[0062] At the end of this classification step, the vectors V[D_i] are subject to a symmetry operation consisting of assigning the probability values of a class corresponding to the anatomical zone of a hand, for example the right hand, to the corresponding class of the other hand, for example the left hand; the vector thus obtained is noted S(V[D_i]). As an illustrative example, for a plurality of classes comprising the classes “dactylogram of the palm of the right hand” and “dactylogram of the palm of the left hand”, the probability of belonging to the class “dactylogram of the palm of the right hand” is assigned to the class “dactylogram of the palm of the left hand” and vice versa.

[0063] The sensitivity function can then be expressed in the form of a Euclidean distance between the vector Vs[sD_i] and the vector S(V[D_i]), possibly affected by a weighting factor. Integrated into the previous cost function, the sensitivity function can be written: / = -'LJp(j)iogq(j) +«l2(Ks[j>d7], s( ))

[0064] When training the convolutional artificial neural network, this distance is minimized via the cost function. For this purpose, an Adam-type optimization algorithm can be used.

[0065] According to certain preferred embodiments, the convolutional network CNN of artificial neurons is trained using a plurality of sets of training dactylograms, each set comprising dactylograms of identical size. Thus, the convolutional network of artificial neurons is not influenced, during its training, by possible variations in the dimensions of the dactylograms within the same set. The modeling of the relevant characteristics of the dermatoglyphs represented by the dactylograms is then more precise. The dimensions of the dactylograms between the sets may nevertheless be different, in particular when the method is adapted to take as input data dactylograms of any size, in accordance with certain embodiments described previously.

[0066] According to certain embodiments, the convolutional network CNN of artificial neurons is trained on sets of training dactylograms having been previously augmented by dactylograms selected from said sets at one or more previously defined frequencies, said selected dactylograms having previously been the subject of an amputation and / or trans- operation. formation by axial symmetry.

[0067] The addition of amputated dactylograms to the training sets is particularly advantageous in that it makes it possible to reduce the sensitivity of the method to the possible absence of certain elements of the dactylograms, for example to the absence of one or more fingers when it is a hand that has undergone surgical amputation. The operation of amputating the dactylograms may, for example, consist of the voluntary removal of certain fingers or certain regions of a dactylogram selected from a training set. It may be carried out at the frequency 1 / 6, that is to say that within the training set one dactylogram out of six is selected and is subject to an amputation operation. The new dactylogram thus obtained is then added to the training set.

[0068] The addition of dactylograms having been the subject of an axial symmetry operation is advantageous in that it makes it possible to improve the sensitivity of the method to chirality and to harmonize, within a training set, the distribution of dactylograms between those corresponding to the right hand and those corresponding to the left hand. The axial symmetry operation may, for example, consist of carrying out an axial symmetry operation on a dactylogram selected from a training set, and the new dactylogram thus obtained is then added to said set, its membership class having been previously modified in accordance with the left or right hand that it represents. This operation can be carried out at a frequency 1 / 2, that is to say that within the training set one dactylogram out of two is selected and is the subject of an axial symmetry operation.

[0069] The training fingerprint sets may also be augmented using other known methods such as scaling, contrast changing, rotation operations, and padding operations.

[0070] The method according to the first aspect of the invention is implemented by computer. With reference to [Fig.5], in a second aspect of the invention, there is provided a data processing device 5000 comprising means for implementing a method 3000 according to any one of the embodiments of the first aspect of the invention.

[0071] An example of a device may be a device responsible for automatically executing sequences of arithmetic or logical operations to perform tasks or actions. This device, also called a computer, may comprise one or more central processing units (CPUs) and / or one or more graphics processors (GPUs) 5001 as well as at least one control device adapted to the execution of these operations. It may also comprise other electronic components such as input / output interfaces 5002, non-volatile or volatile storage devices 5003, and communication buses for transferring data between internal components of the device or with external components. One of the input / output devices 5002 may be a user interface for human-machine interaction, e.g., a graphical user interface for displaying human-understandable information.

[0072] According to a third aspect of the invention, there is provided a computer program 15003 comprising instructions which, when the program is executed by a computer, cause the computer to implement a method (3000) according to any one of the embodiments of the first aspect of the invention.

[0073] Any type of programming language, compiled or interpreted, can be used to implement the steps of the method of the invention. The computer program can be part of a software solution, i.e. a collection of executable instructions, codes, scripts or others and / or databases.

[0074] According to a fourth aspect of the invention, there is provided a computer-readable recording medium 5003 comprising instructions which, when executed by a computer, cause the computer to implement a method according to any one of the embodiments of the first aspect of the invention.

[0075] The computer-readable storage medium 5003 is preferably a non-volatile memory, for example a hard disk or a solid-state drive. It may be a removable storage medium or a non-removable storage medium forming part of a computer.

[0076] The computer-readable recording medium 5003 may also be volatile memory within a removable medium. This may facilitate deployment of the invention in many production sites.

[0077] The computer-readable recording medium 5003 may be part of a computer used as a server from which executable instructions may be downloaded and, when executed by a computer, cause the computer to execute a method according to one of the embodiments described herein.

[0078] The computer program 15003 and the medium 5003 on which it is recorded may be implemented in a distributed computing environment, for example cloud computing. The instructions may be executed on a server to which one or more client computers may connect and provide encoded data as input data to a method according to any one of the embodiments of the first aspect of the invention. Once the data has been processed, the result may be downloaded and decoded to the client computer or sent directly, for example, in the form of instructions.

[0079] In a fifth aspect of the invention, there is provided a classification system for fingerprints from a plurality of membership classes, said system comprises: - a device (1000) for acquiring fingerprints; - a data processing device (5000) according to the second aspect of the invention, said data processing device (5000) is further configured to receive and process dactylograms D acquired by the dactylogram acquisition device (1000).

[0080] According to certain embodiments, the data processing device (5000) is further configured to emit a warning signal when a fingerprint acquired by the acquisition device does not conform to the class expected for said fingerprint and / or is assigned to a non-compliant class by an operator of the system. In particular, the device can be configured according to any one of the three examples of implementation of the method according to the first aspect of the invention described above. Example

[0081] In order to illustrate the performance of the method according to the invention, an example of a method is provided in accordance with certain embodiments described previously. The convolutional network implemented is that described in relation to [Fig.4]. It was trained on a training set comprising more than 3000 fingerprints classified according to the 11 membership classes such as those illustrated in [Fig.2],

[0082] In [Fig.6] are represented the occurrences, expressed as a percentage, of correct (broken line) and incorrect (continuous line) classification of the dactylograms as a function of the P_K probability values of membership inferred by the method according to the example. The graph shows that 95% of the dactylograms that were correctly classified were so classified with a probability greater than 0.9. And only 15% of the dactylograms that were incorrectly classified were so classified with a probability greater than 0.9.

[0083] By setting, for the probability of belonging, a reliability threshold of 0.9, below which the prediction is considered unreliable, 5.4% of the dactylograms in the training set can be eliminated, and the percentage of dactylograms correctly classified out of the 94.6% of the remaining dactylograms reaches 99.7%. It was found that the 5.4% of the dactylograms in the training set thus eliminated corresponded to so-called difficult cases for which the attribution of a class of belonging is not easy even subject to the expertise of a human operator.

[0084] In use, inference times for a fingerprint provided to a method according to the present example are of the order of milliseconds for a convolutional network of 1600 bytes to 400000 parameters stored in 32 bits running on an Intel® Core™ i7-7700 CPU. References Patent literature

[0085] US 5572597 Al, LORAL CORP [US], 05.11.1996.

[0086] EP 0 779 595 A2, NEC CORP [JP], 18.06.1997.

[0087] US 5 825 907 Al, LUCENT TECHNOLOGIES INC [US], 20.10.1998.

[0088] US 2012 014569 Al, IB KOREA LTD [KR], 19.01.2012.

[0089] US 9 530 042 Bl, UNIV KING SAUD [SA], 27.12.2016.

[0090] US 2017 046554 Al, NEC CORP [JP], 16.02.2017.

[0091] EP 3 825 915 Al, IDEMIA IDENTITY & SECURITY FRANCE [FR], 26.05.2021. Littérature non-brevet

[0092] F. Galton, Fingerprint Directories. London, MacMillan & Co, 1895.

[0093] Henry Faulds, Guide to fingerprint Identification, Tokyo, Hanley, 1905.

[0094] E. Henry, Classification and uses of finger prints, published by his majesty’s stationery office, London, 1913.

[0095] Wang et al., Fingerprint Classification Based on Depth Neural Network, arXiv preprint arXiv: 1409.5188, 2014.

[0096] Michelsanti et al., Fast Fingerprint Classification with Deep Neural Networks, VISIGRAPP 2017.

Claims

Claims

1. A computer-implemented method (3000) for classifying dactylograms from a plurality of membership classes Cj, each of the classes corresponding to a particular anatomical area Z_i of the palmar surface of a hand, said method (2000) takes, as input data (13000), at least one dactylogram D, and provides, as output data (03000), a membership class C or a list L of membership classes C_k of the dactylogram D from the plurality of membership classes Cj, said method (3000) comprises the following steps: - providing (3001) a CNN convolutional network of artificial neurons, said CNN convolutional network being previously trained on a set E of training data composed of a plurality of dactylograms D_i classified according to a plurality of classes C j, each of the classes of which corresponds to a particular anatomical area of a hand, said CNN convolutional network being configured to provide the probabilities of membership P_j of each dactylogram D_i of the set E to each of the classes of the plurality of classes Cj of membership; - inferring (3002), using the trained CNN convolutional neural network, the membership probabilities P_i of the dactylogram D provided as input data (13000) to each of the classes of the plurality of membership classes Cj; - selecting (3003) a membership class C from the membership classes Cj for which the membership probability P of the dactylogram D provided as input data (13000) is the highest from the membership probabilities P_i inferred for said classes C_j, or a list L of a number k of membership classes C_k selected from the plurality of membership classes Cj and sorted according to an increasing or decreasing order of the membership probabilities P_k to said classes C_k inferred for the dactylogram D provided as input data (13000).

2. The method (3000) of claim 1, such that the classes of the plurality of membership classes are selected from the complete right hand (C1), the complete left hand (C2), a lower area of the palm of the right hand (C3), a lower area of the palm of the left hand (C4), an upper area of the right hand (C5), an upper area of the left hand (C6), the right palm of the writer (C7), the writer's left palm (C8), at least two, preferably three, preferably four fingers of the right hand (C9), at least two, preferably three, preferably four fingers of the left hand (CIO), the thumbs of the right and left hands (Cl 1).

3. Method (3000) according to any one of claims 1 to 2, such that the convolutional network CNN of artificial neurons comprises: - a first sequence SI of convolutional layers of artificial neurons; - a second sequence S2 of residual convolutional blocks of artificial neurons; - a third sequence S3 comprising at least one layer of fully connected artificial neurons.

4. Method (3000) according to claim 3, such that the convolutional CNN network of artificial neurons further comprises, between the first sequence SI and the second sequence S2, a step of spatial resampling, preferably by bilinear interpolation.

5. Method (3000) according to any one of claims 1 to 4, such that the cost function used when training the CNN convolutional network of artificial neurons comprises a cross-entropy type cost function.

6. A method (3000) according to any one of claims 1 to 5, such that the cost function used when training the CNN convolutional network of artificial neurons comprises an axial symmetry sensitivity function.

7. A method (3000) according to any one of claims 1 to 6, such that the convolutional artificial neural network CNN is trained using a plurality of training fingerprint sets, each set comprising fingerprints of identical dimension.

8. Method (3000) according to claim 7, such that the convolutional CNN network of artificial neurons is trained on sets of training fingerprints having been previously augmented by fingerprints selected from said sets at one or more previously defined frequencies, said selected fingerprints having been subject to an operation of amputation and / or transformation by axial symmetry.

9. Data processing device (5000) comprising means for implementing a method (3000) according to any one of the claims- indications 1 to 8.

10. A computer program (15003) comprising instructions which, when the program is executed by a computer, cause the computer to implement a method (3000) according to any one of claims 1 to 8.

11. A computer-readable medium (5003) comprising instructions which, when executed by a computer, cause the computer to implement a method (3000) according to any one of claims 1 to 8.

12. System for classifying dactylograms among a plurality of classes Cj of membership, each of the classes corresponding to a particular anatomical zone Z_i of the palmar surface of a hand, said system comprises: - a device (1000) for acquiring dactylograms; - a data processing device (5000) according to claim 9, said data processing device (5000) is further configured to receive and process dactylograms acquired by the device (1000) for acquiring dactylograms.

13. System according to claim 12, such that the data processing device (5000) is further configured to emit a warning signal when a fingerprint acquired by the acquisition device does not conform to the class expected for said fingerprint and / or is assigned to a non-compliant class by an operator of the system.

14. Use of a method according to any one of claims 1 to 8 for verifying the integrity of a database of fingerprints of different classes.

Images