Cross-age face recognition method and device

Abstract

The embodiment of the present application provides a cross-age face recognition method and device, including determining a face feature vector, and decomposing the face feature vector into a first feature parameter and a second feature parameter; wherein the first feature parameter is not affected by the change of face age, and the second feature parameter changes with the change of face age; establishing a cross-age face recognition model based on the face feature vector, the first feature parameter and the second feature parameter; training the cross-age face recognition model based on the face feature vector and age corresponding to the face image in a cross-age face database to obtain a target cross-age face recognition model; inputting the face feature vector of the face image to be recognized into the target cross-age face recognition model, it can be determined whether there is a target face image matching the face image to be recognized in the cross-age face database, as well as the identification information of the face image to be recognized, thereby effectively improving the recognition accuracy of cross-age face recognition.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of face recognition technology, and in particular, to a method and device for cross-age face recognition. Background Art

[0002] As an important biometric recognition technology, face recognition has been increasingly widely used.

[0003] In the actual application of face recognition, the face image to be recognized is affected by various external factors, such as lighting, occlusion, posture, expression and age, among which the changes in face images due to age changes are particularly prominent. Although the accuracy of general face recognition has been significantly improved with the development of deep learning technology, cross-age face recognition is still very challenging. Since cross-age face recognition has irreplaceable value in many scenarios, such as searching for lost children or long-term wanted criminals, face comparison at long time intervals, etc., cross-age face recognition has become a technical issue that has received much attention in the field of face recognition research.

[0004] At present, traditional cross-age face recognition generally builds recognition models based on manually defined discriminants or experience, but excessive reliance on human experience makes the recognition accuracy of the recognition model not high. Summary of the invention

[0005] The embodiments of the present application provide a cross-age face recognition method and device, which can solve the technical problem of low recognition accuracy of existing cross-age face recognition methods.

[0006] In a first aspect, an embodiment of the present application provides a method for training a cross-age face recognition model, the method comprising:

[0007] Determine a facial feature vector, and decompose the facial feature vector into a first feature parameter and a second feature parameter; wherein the first feature parameter is not affected by the age of the face, and the second feature parameter changes with the age of the face;

[0008] Establishing a cross-age face recognition model according to the face feature vector, the first feature parameter, and the second feature parameter;

[0009] Acquire a face image in a cross-age face database, and train the cross-age face recognition model according to the face feature vector and age corresponding to the acquired face image to obtain a target cross-age face recognition model;

[0010] Among them, the facial images in the cross-age face database are facial images of multiple users at different ages, and the target cross-age face recognition model is used to determine the target facial image in the cross-age face database that matches the facial image to be recognized based on the facial feature vector of the facial image to be recognized, and determine the identification information of the facial image to be recognized.

[0011] In a possible design, the facial feature vector is decomposed into a first feature parameter and a second feature parameter, including:

[0012] Based on a spiking neural network, the facial feature vector is decomposed into a first feature parameter and a second feature parameter.

[0013] In a possible design, the facial feature vector is decomposed into a first feature parameter and a second feature parameter based on a spiking neural network, including:

[0014] Each element in the initial face feature vector is input into the spiking neural network as an input pulse of the spiking neural network, and the first feature parameter and the second feature parameter are determined according to the output result of the spiking neural network.

[0015] In a possible design, the cross-age face recognition model is trained according to the face feature vector and age corresponding to the acquired face image to obtain a target cross-age face recognition model, including:

[0016] According to the acquired facial feature vector corresponding to the facial image and the age, the cross-age face recognition model is trained to determine a feature change function between the second feature parameter and the age;

[0017] The target cross-age face recognition model is determined according to the feature change function.

[0018] In a second aspect, an embodiment of the present application provides a cross-age face recognition method, the method comprising:

[0019] Obtaining a face image to be recognized;

[0020] A target cross-age face recognition model is used to determine a target face image in a cross-age face database that matches the face image to be recognized, and identification information of the face image to be recognized is determined.

[0021] In a possible design, the method of using the target cross-age face recognition model to determine a target face image in a cross-age face database that matches the face image to be recognized includes:

[0022] Obtaining a second feature parameter in the face feature vector of the face image to be identified, wherein the second feature parameter changes with the age of the face;

[0023] According to the second feature parameter in the facial feature vector of the facial image to be identified and the feature change function between the second feature parameter and age obtained by training in the target cross-age face recognition model, the second feature parameter of the facial image to be identified when corresponding to different ages is predicted, and according to the prediction result, a target facial image in the cross-age face database that matches the facial image to be identified is determined.

[0024] In one possible design approach, it also includes:

[0025] The age corresponding to the face image to be recognized is determined based on the second feature parameter in the face feature vector of the face image to be recognized and the feature change function between the second feature parameter trained in the target cross-age face recognition model and the age.

[0026] In a possible design, the determining the identification information of the face image to be recognized includes:

[0027] Obtaining a first feature parameter in a face feature vector of the face image to be identified, wherein the first feature parameter is not affected by age changes of the face;

[0028] Based on the first feature parameter in the facial feature vector of the facial image to be identified, identification information of the facial image to be identified is output.

[0029] In a third aspect, an embodiment of the present application provides a cross-age face recognition model training device, the device comprising:

[0030] A preprocessing module, used to determine a facial feature vector, and decompose the facial feature vector into a first feature parameter and a second feature parameter; wherein the first feature parameter is not affected by the age of the face, and the second feature parameter changes with the age of the face;

[0031] A model building module, used for building a cross-age face recognition model according to the face feature vector, the first feature parameter and the second feature parameter;

[0032] A training module, used to obtain a face image in a cross-age face database, and train the cross-age face recognition model according to the face feature vector and age corresponding to the obtained face image to obtain a target cross-age face recognition model;

[0033] Among them, the facial images in the cross-age face database are facial images of multiple users at different ages, and the target cross-age face recognition model is used to determine the target facial image in the cross-age face database that matches the facial image to be recognized based on the facial feature vector of the facial image to be recognized, and determine the identification information of the facial image to be recognized.

[0034] In a fourth aspect, an embodiment of the present application provides a cross-age face recognition device, the device comprising:

[0035] An input module, used to obtain a face image to be recognized;

[0036] A recognition module is used to use a target cross-age face recognition model to determine a target face image in a cross-age face database that matches the face image to be recognized, and to determine identification information of the face image to be recognized; the target cross-age face recognition model is obtained according to the cross-age face recognition model training device provided in the first aspect.

[0037] In a fifth aspect, an embodiment of the present application provides an electronic device, including: at least one processor and a memory;

[0038] The memory stores computer-executable instructions;

[0039] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the cross-age face recognition model training method provided in the first aspect; or, the at least one processor executes the cross-age face recognition method provided in the second aspect.

[0040] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores computer execution instructions. When a processor executes the computer execution instructions, it implements the cross-age face recognition model training method provided in the first aspect; or, implements the cross-age face recognition method provided in the second aspect.

[0041] In the seventh aspect, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the cross-age face recognition model training method provided in the first aspect; or, implements the cross-age face recognition method provided in the second aspect.

[0042] The cross-age face recognition method and device provided in the embodiments of the present application decompose the first feature parameter that is not affected by the face age change and the second feature parameter that changes with the face age in the face image to be recognized, and then train the above-mentioned cross-age face recognition model according to the face feature vector and age corresponding to the face image in the cross-age face database, and determine the corresponding relationship between the second feature parameter and the face age in the target cross-age face recognition model. Therefore, the target cross-age face recognition model can accurately predict the change of the face features in the face image to be recognized with age, thereby accurately identifying the target face image in the cross-age face database that matches the face image to be recognized, and determining the identification information of the face image to be recognized according to the first feature parameter in the face image to be recognized, thereby effectively improving the recognition accuracy of cross-age face recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments of the present application or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 A flowchart of a cross-age face recognition model training method provided in an embodiment of the present application;

[0045] Figure 2 Schematic diagram of the process of the cross-age face recognition method provided in the embodiment of the present application Figure 1 ;

[0046] Figure 3 Schematic diagram of the process of the cross-age face recognition method provided in the embodiment of the present application Figure 2 ;

[0047] Figure 4 A schematic diagram of program modules of a cross-age face recognition model training device provided in an embodiment of the present application;

[0048] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their common and usual meanings.

[0051] In addition, the terms "include" and "have" and any variations thereof are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to those components expressly listed but may include other components not expressly listed or inherent to such products or devices.

[0052] The term "module" as used in this application refers to any known or later developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0053] Cross-age face recognition has irreplaceable value in many scenarios, such as searching for lost children or long-wanted criminals, face comparison at long time intervals, etc. Therefore, cross-age face recognition has become a highly concerned technical issue in the field of face recognition research.

[0054] Age, as an important attribute of the face, has a significant impact on the reliability of the recognition results. Face aging is a complex nonlinear internal change. Facial features will undergo significant changes in shape and texture with age. Different people have similar changes at the same age, while the same person has different changes at different ages. Therefore, face aging modeling is a very complex process. In view of these difficulties, in some implementations, the following methods are mainly used:

[0055] 1. Researchers use experience to build facial age estimation algorithms and simulation generation models for faces of different age groups.

[0056] 2. Through discriminant model extraction, including Histogram of Oriented Gradient (HOG), Scale-invariant feature transform (SIFT), Speeded Up Robust Features (SURF), Oriented Fast and Rotated Brief (ORB), Local Binary Patterns (LBP), HAA and other traditional visual feature extraction algorithms, cross-age information is encoded based on experience.

[0057] 3. Improved algorithms based on discriminant models, such as raster scanning.

[0058] 4. Improved algorithms based on Convolutional Neural Networks (CNN), such as CNN-based feature learning and distance measurement learning algorithms.

[0059] 5. Design of generative model and discriminative model based on GAN generative adversarial network.

[0060] Among them, the above 1, 2, and 3 are based on manually defined discriminants or empirical analysis to build algorithms. The extracted features are not invariant enough and are easily disturbed by facial changes caused by age changes. In addition, the excessive reliance on researchers' experience limits the generalization ability of the model. Although the above 2 and 3 use discriminant models to partially improve the shortcomings of pure empirical construction, pure discriminants still find it difficult to capture all information, resulting in low recognition rates. 4 is based on CNN. Although it improves the stability of feature extraction, pure CNN cannot completely solve the problem of extracting invariants of faces that change over time, and still needs to rely on researchers' empirical predictions to simulate the growth process.

[0061] Algorithms based only on 1, 2, 3, and 4 cannot solve the individual differences caused by aging, especially when the age span is large, the difference in facial information of the same individual at different ages may even be greater than the difference in facial information between individuals. Item 5 is based on a generative adversarial network. First, an efficient generative model is designed to generate facial images of different ages to help with recognition. However, the synthetic faces generated by this adversarial network cannot guarantee consistency with natural aging, so it will affect the accuracy when recognizing real faces.

[0062] In order to solve the above technical problems, the embodiment of the present application provides a cross-age face recognition method, which can accurately predict the change of facial features in the face image to be recognized with age by decomposing non-age feature parameters and age feature parameters in the face image to be recognized, and according to the correspondence between the determined age feature parameters and the face age, so as to accurately recognize the target face image that matches the face image to be recognized in the cross-age face database, as well as the identity information of the face image to be recognized. The following is a detailed description using a detailed embodiment.

[0063] Reference Figure 1 , Figure 1 A schematic diagram of a cross-age face recognition model training method provided in an embodiment of the present application. In a feasible implementation, the cross-age face recognition model training method includes:

[0064] S101. Determine a facial feature vector, and decompose the facial feature vector into a first feature parameter and a second feature parameter.

[0065] Among them, the above-mentioned first characteristic parameter is not affected by the change of face age, and the second characteristic parameter changes with the change of face age.

[0066] It is understood that different faces have different dimensions, similar faces have similar dimensions, and the number representing a face (or an element in the training set) is called a face feature vector. A face feature vector includes various numbers in a specific order.

[0067] For example, after mapping a face to a facial feature vector, the facial feature vector can be composed of the following different features:

[0068] Face length (cm)

[0069] Width of face (cm)

[0070] Average skin color of the face (R, G, B)

[0071] Lip width (cm)

[0072] Nose length (cm)

[0073] When a face image is given, different features can be annotated and converted into a face feature vector as follows: (23.1, 15.8, 255, 224, 189, 5.2, 4.4), which means that the length of the face is 23.1 cm, the width of the face is 15.8 cm, the average skin color of the face is (255, 224, 189), the lip width is 5.2 cm, and the nose length is 4.4 cm.

[0074] In other embodiments, other features may be derived from the facial image, such as hair color, beard, glasses, etc.

[0075] In some embodiments, facial images from a cross-age face database may be acquired, and feature extraction preprocessing may be performed on the acquired facial images. Then, facial feature vectors may be determined based on a feature extraction network.

[0076] Optionally, the above cross-age face database can be the MORPH face database. The academic version of the MORPH face database has 55,134 images and 13,618 people, with an average of only four images per person, ranging in age from 16 to 77 years old. The age span of each person is relatively small, and contains large posture and expression interference.

[0077] In a feasible implementation manner, feature extraction preprocessing is performed on the acquired face image, including matrixing and graying the face image.

[0078] Optionally, the acquired facial images can be uniformly scaled to images with a resolution of 400*500. In this process, if the original facial image is not in a 4:5 ratio, black pixels are filled in the blanks, that is, RGB=(0, 0, 0) in the blanks; then, the color information in the acquired facial images is discarded, that is, the RGB three-channel color image is converted into a grayscale image.

[0079] Optionally, the following empirical formula can be used to convert the RGB three-channel color image into a grayscale image:

[0080] Gray value = red channel (R) * 0.299 + green channel (G) * 0.587 + blue channel (B) * 0.114

[0081] Since the human eye is extremely sensitive to brightness information, this empirical formula retains the brightness information to the greatest extent according to the color ratio, generates a grayscale image that best matches the human eye's vision, and saves computing resources. In this way, a single-channel 400*500 two-dimensional matrix is ​​obtained. Subsequently, for this matrix, the Histogram of Oriented Gradient (HOG) algorithm can be used to calibrate the various feature positions of the face. This step has five key points, including the center of the two eyes, the tip of the nose, and the two corners of the mouth. Based on the information of these five key points, the facial information can be determined (that is, the specific position of the face can be located).

[0082] Furthermore, based on the above facial information, the original face can be divided into multiple image areas. Here, with the above five key points as the center, five squares of size 100*100 are selected as five feature areas. At the same time, based on the whole face, the whole face, upper 3 / 4 part, 1 / 16 to 13 / 16 part, 1 / 8 to 3 / 4 part, and lower 3 / 4 part are extracted, a total of five feature areas, to compensate for the recognition errors that may occur in the samples of cross-face alignment.

[0083] In the above way, a total of 10 different image regions are obtained. These 10 image regions can be compressed into 50*50 dimensions to reduce the amount of calculation. In addition, these 10 different image regions can also be expanded into 40 different image regions through rotation scale transformation (clockwise rotation of 90, 180, 270 degrees). These 40 different 50*50 image regions can be used as input for the next step of feature vector extraction.

[0084] In the embodiment of the present application, a feature extraction network can be used to extract facial feature vectors from the above 40 different image areas of 50*50 size.

[0085] Optionally, the feature extraction network may include three convolutional layers, three maximum pooling layers (including the following ReLU nonlinear activation function) and a final fully connected layer, with the input being the 40 different image regions of 50*50 size obtained above.

[0086] In some embodiments, the architecture of the feature extraction network is as follows:

[0087] Convolutional layer 1: Use four 5*5 convolution kernels to convolve the 50*50 image and output 46*46*4 to the next layer.

[0088] Max pooling layer 1: Use 2*2 max pooling unit for pooling, and use ReLU function to add nonlinear part, output 23*23*4 to the next layer.

[0089] Convolutional layer 2: Use four 4*4 convolution kernels to convolve the 23*23 image and output 20*20*16 to the next layer.

[0090] Max pooling layer 2: Use 2*2 max pooling unit for pooling, and use ReLU function to add nonlinear part, output 10*10*16 to the next layer.

[0091] Convolutional layer 3: Use four 3*3 convolution kernels to convolve the 10*10 image and output 8*8*64 to the next layer.

[0092] Max pooling layer 3: Use 2*2 max pooling unit for pooling, and use ReLU function to add nonlinear part, output 4*4*64 to the fully connected layer.

[0093] Fully connected layer: The 64 sets of feature matrices based on different convolution kernels obtained above are vectorized to obtain 64 sets of 1*16 column vectors.

[0094] In the embodiment of the present application, the above 64 groups of 1*16 column vectors can be used as face feature vectors.

[0095] Since a face contains inherent identity information and age information, face features can be represented by identity-related features and age-related features. Based on this, the face feature vector can be decomposed into these two unrelated parts, that is, the face feature vector can be decomposed into a first feature parameter and a second feature parameter.

[0096] For example, since the name corresponding to a face is not affected by the age of the face, the name of the face can be determined as the first feature parameter, while features such as face length, face width, lip width, and nose length may change with the age of the face, so these parameters can be used as second feature parameters.

[0097] S102: Establish a cross-age face recognition model based on the facial feature vector, the first feature parameter and the second feature parameter.

[0098] In the embodiment of the present application, a cross-age face recognition model is constructed. Assuming that the face feature vector is x, it can be defined as:

[0099] x=x inv +x(t) age

[0100] where x inv is the first characteristic parameter, which is not affected by the age change of the face. age is the second characteristic parameter, which changes with the face age (t). In the above cross-age face recognition model, x(t) age It can be a discrete mapping with steps of years (t).

[0101] S103, obtaining a face image from a cross-age face database, and training the cross-age face recognition model according to the face feature vector and age corresponding to the obtained face image to obtain a target cross-age face recognition model.

[0102] In an embodiment of the present application, the above-mentioned cross-age face database can be trained based on the facial feature vector and facial age information of the acquired facial image to determine the feature change function between the second feature parameter and the facial age.

[0103] When performing cross-age face recognition, based on the second feature parameter of the face image to be recognized and the feature change function between the predetermined second feature parameter and the face age, it is possible to accurately predict how the facial features of the face image to be recognized change with age. Then, based on the prediction result, the face recognition image to be recognized is matched with each face image in the above-mentioned cross-age face database to find the target face image that matches the face image to be recognized.

[0104] In addition, identification information of the face image to be identified may also be identified based on the first characteristic parameter in the face image to be identified.

[0105] For example, when the first characteristic parameter is a name, the identity information of the subject can be determined therefrom.

[0106] The cross-age face recognition model training method provided in the embodiment of the present application decomposes the first feature parameter that is not affected by the face age change and the second feature parameter that changes with the face age in the face image to be recognized, and then trains the above-mentioned cross-age face recognition model according to the face feature vector and age corresponding to the face image in the cross-age face database, and determines the corresponding relationship between the second feature parameter and the face age in the target cross-age face recognition model. Therefore, the target cross-age face recognition model can be used to accurately predict the change of facial features in the face image to be recognized with age, thereby accurately identifying the target face image in the cross-age face database that matches the face image to be recognized, and determining the identification information of the face image to be recognized according to the first feature parameter in the face image to be recognized, thereby effectively improving the recognition accuracy of cross-age face recognition.

[0107] Based on the contents described in the above embodiments, in a feasible implementation of the present application, a spiking neural network can be used to accurately decompose the facial feature vector x into the first feature parameter x inv and the second characteristic parameter x(t) age Two parts.

[0108] Limited by the network structure, in the training of the pulse neural network, the initial facial feature vector cannot be a continuous numerical form similar to the traditional method, but must first be converted into a time-dependent pulse form. Through the above facial feature vector extraction, 64 groups of 1*16 column vectors are obtained. Here, the 16 elements in each column vector are converted into input pulses and evenly distributed on the time axis with an interval of 10ms. The intensity (amplitude) of the pulse is the numerical size of the element. In this way, 64 pulse neurons are constructed as the input layer.

[0109] In this embodiment, the spiking neural network simulates the propagation of neural signals by simulating the transmission of neuronal synaptic potentials, and then simulates the decision-making process of the human nervous system. The spiking locally competitive algorithms (S-LCA) are used to simulate this process, and the expression of synaptic potential is defined as:

[0110]

[0111] Among them, b i is the initial potential of the external input, ω ij To suppress the signal strength weight, when t is greater than 0, α(t) = e -t , otherwise α(t)=0,σ i (t) is the synaptic potential sequence, defined as σ i (t) = ∑δ(tt i，k ), where t i,k is an ordered time series, δ(t) is a simulated synaptic sequence, and * represents a convolution operation.

[0112] The neurons constructed based on the above algorithm can well simulate the physiological potential of synaptic transmission, thereby achieving the effect of time-efficient and efficient training.

[0113] In some embodiments, the architecture of the spiking neural network may be:

[0114] After the input layer, there are 64 fully connected layers (S-LCA), and the synapses of neurons in each fully connected layer are followed by a factorization layer, which also contains 64 neurons, corresponding one to one with the input layer.

[0115] The decomposition of the age factor is mainly achieved by comparing the pulse intensity changes and time differences of the input and output layers: all pulses are divided into 16 blocks according to time, corresponding to the 16 input column vectors mentioned above, and then the average value in each block is taken as the main pulse time unknown, and all pulses are divided according to the intensity and the time difference with the main pulse to establish an attenuation model: A = e -Δt ·A0, convert the equivalent main pulse weight, and finally add all weights to the main pulse. In this way, the time difference and intensity difference of the input and output pulses can be compared. If the relative time difference or intensity change of the pulse is greater, the age factor of the input vector will account for a higher proportion. Conversely, within a smaller time difference, the greater the intensity change, the higher the age factor of the input vector. age The faster the rising rate, the greater the specific proportion and x(t) age The function fitting of can be calculated by the following formula:

[0116] x(t) age =1.44 Δt +1.1ΔA ·t

[0117] In order to further separate the age factor, the above factor decomposition layer decomposes x(t) age The part will continue to be used as input and fed into the same 64-layer fully connected and a factor decomposition layer to further extract x(t) age The possible non-age factors in the t are decomposed into x(t) by using a softmax fully connected layer and calculating the weight x according to the above proportion. age and x inv Afterwards, two scores are calculated, corresponding to age changes and identity characteristics, and used as output. Finally, they are compared with the annotations in the training data set to determine whether the training is accurate.

[0118] The identity identification process directly calls x inv , based on CosFace module discrimination. Use this neural network to obtain x(t) age With x inv The correlation is low enough that there is no need to add a decorrelation algorithm.

[0119] For a better understanding of the embodiments of the present application, refer to Figure 2 , Figure 2 Schematic diagram of the process of the cross-age face recognition method provided in the embodiment of the present application Figure 1 .

[0120] In an embodiment of the present application, after a face image is input, an initial face feature vector of the face image is extracted, and then the initial face feature vector is decomposed into a second feature parameter (including part of the first feature parameter) and a first feature parameter based on a pulse neural network 1. After that, the decomposed second feature parameter (including part of the first feature parameter) is further decomposed into a second feature parameter and a first feature parameter based on a pulse neural network 2.

[0121] The cross-age face recognition method provided in the embodiment of the present application can decompose the first feature parameter and the second feature parameter in the face feature vector model by utilizing the training of the pulse neural network. By training the cross-age face recognition model, the feature change function between the second feature parameter and the age can be obtained, so that the above-mentioned cross-age face recognition model has the ability to predict the change of face feature data according to age, which can effectively improve the recognition accuracy of cross-age face recognition.

[0122] Based on the contents described in the above embodiments, Figure 3 , Figure 3 Schematic diagram of the process of the cross-age face recognition method provided in the embodiment of the present application Figure 2 In a feasible implementation of the present application, the above-mentioned cross-age face recognition method includes:

[0123] S301: Obtain a face image to be recognized.

[0124] In this embodiment, after the face image to be recognized is obtained, the face feature vector of the face image to be recognized is determined and input into the target cross-age face recognition model.

[0125] S302: using the target cross-age face recognition model, determine a target face image in a cross-age face database that matches the face image to be recognized, and determine identification information of the face image to be recognized.

[0126] In an embodiment of the present application, when performing cross-age face recognition, based on the second feature parameter of the face image to be recognized and the feature change function between the second feature parameter and age obtained by training in the target cross-age face recognition model, it is possible to accurately predict how the facial features of the face image to be recognized change with age, and then, based on the prediction result, the face recognition image to be recognized is matched with each face image in the cross-age face database, so that the target face image that matches the face image to be recognized can be found.

[0127] In a feasible implementation manner of the present application, after extracting the first feature parameter in the face image to be identified through the face feature vector model, identification information of the face image to be identified can also be determined based on the extracted first feature parameter.

[0128] The cross-age face recognition method provided in the embodiment of the present application can accurately predict how the facial features in the face image to be recognized change with age by decomposing the first feature parameter and the second feature parameter in the face image to be recognized, and based on the feature change function between the determined second feature parameter and the face age, thereby accurately identifying the target face image that matches the face image to be recognized in the cross-age face database, as well as the identification information of the face image to be recognized, effectively improving the recognition accuracy of cross-age face recognition.

[0129] Based on the contents described in the above embodiments, in a feasible implementation manner of the present application, in the above cross-age face recognition method, a cross-age face recognition model can be first constructed. Assuming that the face feature vector is x, it can be defined as:

[0130] x=x inv +x(t) age

[0131] where x inv is the first characteristic parameter, which is not affected by the age change of the face. age is the second feature parameter, which changes with the face age (t). In the above face feature vector model, x(t) age Can be a discrete mapping with year steps.

[0132] Obtain the face images in the cross-age face database, and train the cross-age face recognition model based on the face feature vector and face age information of the obtained face images to determine the second feature parameter x(t) age The feature change function between face age (t).

[0133] A face image to be identified is obtained, a face feature vector of the face image to be identified is determined, and the face feature vector of the face image to be identified is input into the above-trained cross-age face recognition model.

[0134] The cross-age face recognition model is used to extract the second feature parameters in the face image to be recognized. According to the extracted second feature parameters and the above-mentioned feature change function obtained through training, the second feature parameters of the face image to be recognized corresponding to different ages are predicted. According to the prediction result, the recognized face image is matched with each face image in the cross-age face database to determine the target face image that matches the face image to be recognized, which can effectively improve the recognition accuracy of cross-age face recognition.

[0135] Furthermore, in an embodiment of the present application, after determining the target facial image that matches the facial image to be identified, the age information corresponding to the facial image to be identified can also be determined based on the second feature parameter in the facial feature vector of the facial image to be identified and the above-mentioned trained feature change function.

[0136] Based on the contents described in the above embodiments, the present application also provides a cross-age face recognition model training device. Figure 4 , Figure 4 Schematic diagram of a program module of a cross-age face recognition model training device provided in an embodiment of the present application. The cross-age face recognition model training device 40 includes:

[0137] The preprocessing module 401 is used to determine a facial feature vector and decompose the facial feature vector into a first feature parameter and a second feature parameter; wherein the first feature parameter is not affected by the age change of the face, and the second feature parameter changes with the age change of the face.

[0138] The model building module 402 is used to build a cross-age face recognition model according to the facial feature vector, the first feature parameter and the second feature parameter.

[0139] The training module 403 is used to obtain face images in the cross-age face database, and train the cross-age face recognition model according to the face feature vectors and ages corresponding to the obtained face images to obtain a target cross-age face recognition model.

[0140] Among them, the facial images in the above-mentioned cross-age face database are facial images of multiple users at different ages, and the target cross-age face recognition model is used to determine the target facial image in the cross-age face database that matches the facial image to be recognized based on the facial feature vector of the facial image to be recognized, and determine the identification information of the facial image to be recognized.

[0141] The cross-age face recognition model training device 40 provided in the embodiment of the present application decomposes the first feature parameter that is not affected by the face age change and the second feature parameter that changes with the face age in the face image to be recognized, and then trains the above-mentioned cross-age face recognition model according to the face feature vector and age corresponding to the face image in the cross-age face database, and determines the correspondence between the second feature parameter and the face age in the target cross-age face recognition model. Therefore, the target cross-age face recognition model can be used to accurately predict the change of the facial features in the face image to be recognized with age, thereby accurately identifying the target face image in the cross-age face database that matches the face image to be recognized, and determining the identification information of the face image to be recognized according to the first feature parameter in the face image to be recognized, thereby effectively improving the recognition accuracy of cross-age face recognition.

[0142] In a possible design, the preprocessing module 401 is specifically used to decompose the facial feature vector into a first feature parameter and a second feature parameter based on a pulse neural network.

[0143] In one possible design, each element in the initial facial feature vector can be input into the pulse neural network as an input pulse of the pulse neural network, and the first feature parameter and the second feature parameter can be determined based on the output result of the pulse neural network.

[0144] In a possible design, the training module 403 is used to:

[0145] According to the acquired facial feature vector corresponding to the facial image and the age, the cross-age face recognition model is trained to determine a feature change function between the second feature parameter and the age;

[0146] The target cross-age face recognition model is determined according to the feature change function.

[0147] It should be noted that the specific implementation contents of the preprocessing module 401, the model building module 402 and the training module 403 in the embodiment of the present application can be referred to in Figures 1 to 3 The relevant contents in the illustrated embodiment will not be described in detail here.

[0148] Based on the contents described in the above embodiments, a cross-age face recognition device is also provided in the embodiments of the present application. The device includes:

[0149] The input module is used to obtain the face image to be recognized.

[0150] The recognition module is used to use the above-mentioned target cross-age face recognition model to determine the target face image in the cross-age face database that matches the face image to be recognized, and to determine the identification information of the face image to be recognized.

[0151] In a possible design, the identification module is specifically used for:

[0152] A second feature parameter in the face feature vector of the face image to be recognized is obtained, where the second feature parameter changes with the age of the face.

[0153] According to the second feature parameter in the facial feature vector of the facial image to be identified and the feature change function between the second feature parameter and age obtained by training in the target cross-age face recognition model, the second feature parameter of the facial image to be identified when corresponding to different ages is predicted, and according to the prediction result, a target facial image in the cross-age face database that matches the facial image to be identified is determined.

[0154] In a possible design, the identification module is also used to:

[0155] The age corresponding to the face image to be recognized is determined based on the second feature parameter in the face feature vector of the face image to be recognized and the feature change function between the second feature parameter trained in the target cross-age face recognition model and the age.

[0156] In a possible design, the identification module is also used to:

[0157] Obtain a first feature parameter in a facial feature vector of the facial image to be identified, where the first feature parameter is not affected by age changes of the face; and output identification information of the facial image to be identified based on the first feature parameter in the facial feature vector of the facial image to be identified.

[0158] Furthermore, based on the contents described in the above embodiments, an electronic device is also provided in an embodiment of the present application, which includes at least one processor and a memory; wherein the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory to implement the various steps of the cross-age face recognition model training method or the cross-age face recognition method as described in the above embodiments, which will not be repeated in this embodiment.

[0159] In order to better understand the embodiments of the present application, refer to Figure 5 , Figure 5 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application.

[0160] like Figure 5 As shown, the electronic device 50 of this embodiment includes: a processor 501 and a memory 502; wherein:

[0161] Memory 502, used to store computer-executable instructions;

[0162] Processor 501 is used to execute computer-executable instructions stored in the memory to implement the various steps of the cross-age face recognition model training method or the cross-age face recognition method described in the above embodiments. For details, please refer to the relevant description in the above method embodiments.

[0163] Optionally, the memory 502 may be independent or integrated with the processor 501 .

[0164] When the memory 502 is independently provided, the device further includes a bus 503 for connecting the memory 502 and the processor 501 .

[0165] Furthermore, based on the contents described in the above embodiments, a computer-readable storage medium is also provided in the embodiments of the present application, wherein the computer-readable storage medium stores computer-executable instructions. When the processor executes the computer-executable instructions, the various steps of the cross-age face recognition model training method or the cross-age face recognition method described in the above embodiments are implemented.

[0166] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0167] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0168] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The above-mentioned module-composed unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0169] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to perform some steps of the method described in each embodiment of the present application.

[0170] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the application may be directly implemented as being executed by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.

[0171] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0172] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0173] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0174] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0175] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A cross-age face recognition model training method, characterized in that: The method comprises: Determine a facial feature vector, and decompose the facial feature vector into a first feature parameter and a second feature parameter; wherein the first feature parameter is not affected by the age of the face, and the second feature parameter changes with the age of the face; Establishing a cross-age face recognition model according to the face feature vector, the first feature parameter, and the second feature parameter; Acquire a face image in a cross-age face database, and train the cross-age face recognition model according to the face feature vector and age corresponding to the acquired face image to obtain a target cross-age face recognition model; The face images in the cross-age face database are face images of multiple users at different ages, and the target cross-age face recognition model is used to determine the target face image in the cross-age face database that matches the face image to be recognized according to the face feature vector of the face image to be recognized, and to determine the identification information of the face image to be recognized; Decomposing the facial feature vector into a first feature parameter and a second feature parameter includes: Based on a spiking neural network, decomposing the facial feature vector into a first feature parameter and a second feature parameter; The step of training the cross-age face recognition model according to the face feature vector and age corresponding to the acquired face image to obtain a target cross-age face recognition model includes: According to the acquired facial feature vector corresponding to the facial image and the age, the cross-age face recognition model is trained to determine a feature change function between the second feature parameter and the age; The target cross-age face recognition model is determined according to the feature change function.

2. The method according to claim 1, characterized in that The method of decomposing the facial feature vector into a first feature parameter and a second feature parameter based on a spiking neural network includes: Each element in the initial face feature vector is input into the spiking neural network as an input pulse of the spiking neural network, and the first feature parameter and the second feature parameter are determined according to the output result of the spiking neural network.

3. A cross-age face recognition method, characterized in that: The method comprises: Obtaining a face image to be recognized; A target cross-age face recognition model is used to determine a target face image in a cross-age face database that matches the face image to be recognized, and to determine identification information of the face image to be recognized; the target cross-age face recognition model is obtained according to the cross-age face recognition model training method according to any one of claims 1 to 2.

4. The method according to claim 3, characterized in that The method of using the target cross-age face recognition model to determine a target face image in a cross-age face database that matches the face image to be recognized includes: Obtaining a second feature parameter in the face feature vector of the face image to be identified, wherein the second feature parameter changes with the age of the face; According to the second feature parameter in the facial feature vector of the facial image to be identified and the feature change function between the second feature parameter and age obtained by training in the target cross-age face recognition model, the second feature parameter of the facial image to be identified when corresponding to different ages is predicted, and according to the prediction result, a target facial image in the cross-age face database that matches the facial image to be identified is determined.

5. The method according to claim 4, characterized in that Also includes: The age corresponding to the face image to be recognized is determined based on the second feature parameter in the face feature vector of the face image to be recognized and the feature change function between the second feature parameter trained in the target cross-age face recognition model and the age.

6. The method according to claim 3, characterized in that: The determining of identification information of the face image to be identified includes: Obtaining a first feature parameter in a face feature vector of the face image to be identified, wherein the first feature parameter is not affected by age changes of the face; Based on the first feature parameter in the facial feature vector of the facial image to be identified, identification information of the facial image to be identified is output.

7. A cross-age face recognition model training device, characterized in that: The device comprises: A preprocessing module, used to determine a facial feature vector, and decompose the facial feature vector into a first feature parameter and a second feature parameter; wherein the first feature parameter is not affected by the age of the face, and the second feature parameter changes with the age of the face; A model building module, used for building a cross-age face recognition model according to the face feature vector, the first feature parameter and the second feature parameter; A training module, used to obtain a face image in a cross-age face database, and train the cross-age face recognition model according to the face feature vector and age corresponding to the obtained face image to obtain a target cross-age face recognition model; The face images in the cross-age face database are face images of multiple users at different ages, and the target cross-age face recognition model is used to determine the target face image in the cross-age face database that matches the face image to be recognized according to the face feature vector of the face image to be recognized, and to determine the identification information of the face image to be recognized; The preprocessing module is specifically used to decompose the facial feature vector into a first feature parameter and a second feature parameter based on a spiking neural network; The training module is specifically used to train the cross-age face recognition model according to the facial feature vector and age corresponding to the acquired facial image, and determine the feature change function between the second feature parameter and age; based on the feature change function, determine the target cross-age face recognition model.

8. A cross-age face recognition device, characterized in that: The device comprises: An input module, used to obtain a face image to be recognized; A recognition module is used to use a target cross-age face recognition model to determine a target face image in a cross-age face database that matches the face image to be recognized, and to determine identification information of the face image to be recognized; the target cross-age face recognition model is obtained according to the cross-age face recognition model training device according to claim 7.

9. An electronic device, characterized in that: include: at least one processor and memory; The memory stores computer-executable instructions; The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the cross-age face recognition model training method according to any one of claims 1 to 2; Alternatively, the at least one processor is enabled to execute the cross-age face recognition method as described in any one of claims 3 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the processor executes the computer-executable instructions, the cross-age face recognition model training method according to any one of claims 1 to 2 is implemented; Alternatively, implement the cross-age face recognition method as described in any one of claims 3 to 6.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the cross-age face recognition model training method according to any one of claims 1 to 2 is implemented; Alternatively, implement the cross-age face recognition method described in any one of claims 3 to 6.

Images