Face type recognition method, device and equipment for heterogeneous faces and storage medium

Abstract

The application discloses a heterogeneous face type recognition method, device and equipment and a storage medium, and the method comprises the following steps: obtaining a to-be-recognized image; extracting a to-be-recognized feature vector from the to-be-recognized image; comparing the to-be-recognized feature vector with each special face type label feature vector respectively to obtain a target special face type satisfying a preset condition, wherein the special face type is pre-set, and the label feature vector is calculated from image feature vectors extracted from multiple special face images, wherein the multiple special face images belong to multiple special face types; and outputting the target special face type as the face type of the to-be-recognized image. The application compares the image feature vector of the to-be-recognized image with the label feature vector pre-constructed for representing the special face type features to confirm the special face type to which the face image in the to-be-recognized image belongs, and the recognition accuracy is high, the response speed is fast, and the interpretability is strong.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a heterogeneous face type recognition method and device, equipment and storage medium. BACKGROUND

[0002] With the development of artificial intelligence, face recognition has always been a popular field of research in all circles, and has made great progress on the basis of deep convolutional neural networks and is widely used in various industries. For example, in Internet risk control, the content of the face image published by the user needs to be identified to confirm whether there is sensitive information. At present, this review method can be realized through a machine learning model, and some users will use cartoons or sketches to express the outline of the face in order to pass the review of the machine learning model, and exaggerate or deform the features of the characters, so that the machine learning model is difficult to detect the problem, resulting in illegal image data flowing into the Internet and causing negative effects.

[0003] In the detection algorithm, for the face photographed by the camera, face recognition is based on deep convolutional neural network, which is relatively mature in the industry and has been widely applied. At the same time, the construction of sketch / cartoon face recognition model is based on sketch / cartoon samples to extract the bottom features. However, once the above three types of images are mixed together for heterogeneous face image recognition, the features of real faces and sketches / cartoons are not comparable, and traditional face recognition algorithms cannot correctly match the above images. In terms of application resources, most face recognition methods currently use deep neural networks, which use complex calculations to improve model accuracy by building multiple layers of networks, but at the same time, the resource cost is also huge, and if multiple parallel face recognition models are used, the cost will also increase proportionally. SUMMARY

[0004] The present application provides a heterogeneous face type recognition method, device, equipment and storage medium to solve the problem of low accuracy and low efficiency of existing heterogeneous face recognition methods.

[0005] To solve the above technical problems, one technical solution adopted by the present application is to provide a heterogeneous face type recognition method, comprising: obtaining a to-be-identified image; extracting a to-be-identified feature vector from the to-be-identified image; comparing the to-be-identified feature vector with each special face type label feature vector respectively to obtain a target special face type that meets a preset condition, the special face type being pre-set, and the label feature vector being calculated from image feature vectors extracted from multiple special face images, wherein the multiple special face images belong to multiple special face types; and outputting the target special face type as the face type of the to-be-identified image.

[0006] As a further improvement of the present application, the label feature vector calculation process comprises: extracting the image feature vector of each special face image by using the pre-trained deep neural network; calculating the mean feature vector and the standard deviation feature vector of each special face type according to all image feature vectors corresponding to each special face type, and the mean feature vector and the standard deviation feature vector constitute the label feature vector.

[0007] As a further improvement of the present application, the target special face type satisfying the preset condition is obtained by comparing the to-be-identified feature vector with the label feature vector of each special face type, comprising: constructing the feature value range of each dimension corresponding to each special face type by using the feature value of each dimension corresponding to the mean feature vector and the standard deviation feature vector of each special face type; comparing the feature value of each dimension of the to-be-identified feature vector with the feature value range of each dimension corresponding to each special face type to obtain a first special face type satisfying the preset range condition, and the preset range condition comprises that the feature value of each dimension of the to-be-identified feature vector falls within the feature value range of the corresponding dimension of the first special face type; calculating the Euclidean distance between the image feature vector of each special face image in the first special face type and the to-be-identified feature vector; when there is at least one Euclidean distance satisfying the preset distance condition, the first special face type is retained, otherwise the first special face type is deleted; selecting the first special face type corresponding to the image feature vector with the smallest Euclidean distance as the target special face type.

[0008] As a further improvement of the present application, the method further comprises: when there is no first special face type satisfying the preset range condition or there is no Euclidean distance satisfying the preset distance condition, the to-be-identified image is sent to manual review.

[0009] As a further improvement of the present application, the to-be-identified feature vector is extracted from the to-be-identified image, comprising: detecting whether there is a face image in the to-be-identified image by using a pre-trained multi-task convolutional neural network; if there is, segmenting the face image from the to-be-identified image, and then extracting the feature vector of the face image by using a pre-trained deep neural network to obtain the to-be-identified feature vector; if there is not, skipping the current to-be-identified image and continuing to identify the next to-be-identified image.

[0010] As a further improvement of the present application, the to-be-identified feature vector is extracted from the to-be-identified image, comprising: determining whether there are multiple face images in the to-be-identified image; if yes, segmenting at least one face image from the to-be-identified image and assigning a unique identifier to each face image; extracting the image feature of each face image by using a deep neural network to obtain the to-be-identified feature vector of each face image.

[0011] As a further improvement of the present application, the number of special face images corresponding to each special face type is the same, and each special face image is a single face image.

[0012] To solve the above technical problems, another technical solution adopted by the present application is to provide a heterogeneous face type recognition device, comprising: an acquisition module configured to acquire a to-be-identified image; an extraction module configured to extract a to-be-identified feature vector from the to-be-identified image; a comparison module configured to compare the to-be-identified feature vector with each special face type label feature vector respectively, and obtain a target special face type that meets a preset condition, wherein the special face type is pre-set, and the label feature vector is calculated from image feature vectors extracted from a plurality of special face images, wherein the plurality of special face images belong to a plurality of special face types; and an output module configured to output the target special face type as the face type of the to-be-identified image.

[0013] To solve the above technical problems, another technical solution adopted by the present application is to provide a computer device, comprising a processor and a memory coupled to the processor, wherein the memory stores program instructions, and the program instructions are executed by the processor to perform the steps of the heterogeneous face type recognition method according to any one of the above embodiments.

[0014] To solve the above technical problems, another technical solution adopted by the present application is to provide a storage medium storing program instructions capable of implementing the above heterogeneous face type recognition method.

[0015] The beneficial effects of the present application are as follows: the heterogeneous face type recognition method of the present application pre-constructs a label feature vector representing a special face type, extracts a to-be-identified feature vector in a to-be-identified image after obtaining the to-be-identified image, compares the to-be-identified feature vector with each label feature vector respectively, confirms the target label feature vector closest to the to-be-identified feature vector and the target special face type corresponding thereto, and outputs the target special face type as the face type of the to-be-identified image. The method does not need to rely on a multi-layer structure of a deep neural network to achieve recognition of special face images, and has small calculation amount, faster response speed, and less system resources occupied while ensuring accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a flowchart of the heterogeneous face type recognition method of the embodiment of the present application;

[0017] Figure 2 is a functional module diagram of the heterogeneous face type recognition device of the embodiment of the present application;

[0018] Figure 3 is a structural schematic diagram of a computer device according to an embodiment of the present application.

[0019] Figure 4 is a structural schematic diagram of a storage medium according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort fall within the scope of protection of the present application.

[0021] The terms "first", "second", "third" in the present application are only for descriptive purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise explicitly and specifically limited. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative position relationship, movement condition, etc. between the components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0022] In this document, reference to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. The appearances of the phrase "in an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. A person of ordinary skill in the art will understand that an embodiment described herein can be combined with another embodiment.

[0023] Figure 1 is a flowchart of a face type recognition method for a heterogeneous face according to an embodiment of the present application. It should be noted that the method of the present application is not limited to the flow order shown in Figure 1 . As shown in Figure 1 , the method includes the following steps:

[0024] Step S101: Obtain an image to be identified.

[0025] Specifically, with the rapid development of Internet platforms, a large number of users can publish various video and image works through microblogging, TikTok and other means. In order to avoid the spread of image content of sensitive information, it is necessary to audit the video or image works published by users. The image to be identified in the embodiment refers to the video or image work that the user wants to publish. When the user publishes a video work, the video work is cut into images by frame and then identified.

[0026] It should be noted that the face type recognition method of the heterogeneous face in the embodiment is mainly used for type recognition of heterogeneous faces, such as recognition of face images including real faces, cartoon faces and sketch faces, so as to prevent users from publishing illegal image content through heterogeneous faces.

[0027] Step S102: Extract a feature vector to be identified from the image to be identified.

[0028] Specifically, in order to facilitate the identification of the image to be identified by the computer, the embodiment extracts a feature vector to be identified from the image to be identified after obtaining the image to be identified, which is used for subsequent face image recognition. The feature vector of the image can be extracted by using a deep neural network, for example: VGGNet network and Resnet network can be used for feature extraction of the image. In the embodiment, the deep neural network is preferably a Facenet face recognition model.

[0029] In the embodiment, the extracted feature vector of the image includes 128 dimensions, and each dimension corresponds to a feature value.

[0030] Further, it should be understood that not all images in the video or image work published by the user need to be identified. For images that do not include faces, no identification processing is required. Therefore, in order to improve the efficiency of image recognition, step S102 specifically includes:

[0031] 1. Use a pre-trained multi-task convolutional neural network to detect whether there is a face image in the image to be identified.

[0032] Specifically, after obtaining the to-be-recognized image, the to-be-recognized image is input into a multi-task convolutional neural network for detection, and whether the to-be-recognized image contains a face image is detected by using the multi-task convolutional neural network. It should be noted that the multi-task convolutional neural network (Multi-task convolutional neural network, MTCNN) mainly includes three network architectures: P-Net, R-Net, and O-Net. The P-Net is the first layer of the multi-task convolutional neural network, which is used to extract information coordinates that may be a face and input to the R-Net; the R-Net is the second layer of the multi-task convolutional neural network, which is used to extract face data according to the information coordinates input by the P-Net of the previous layer, and then input into a fully connected layer containing 128 neurons to filter out more errors, and finally the output result is optimized by the NMS algorithm; the O-Net is the third layer of the multi-task convolutional neural network, which is used to extract five landmark point information of the face from the output result of the R-Net. Whether the to-be-recognized image contains a face image can be detected by using the multi-task convolutional neural network.

[0033] 2. If yes, a face image is segmented from the to-be-recognized image, and a feature vector of the face image is extracted by using a pre-trained deep neural network to obtain a to-be-recognized feature vector.

[0034] Specifically, if the to-be-recognized image contains a face image, a face image is segmented from the to-be-recognized image, and the segmentation of the face image can be realized by using an RNN model or an FCN model.

[0035] 3. If no, the current to-be-recognized image is skipped and the next to-be-recognized image is continued to be recognized.

[0036] Specifically, when the to-be-recognized image does not contain a face image, it is not necessary to continue to perform face recognition on the to-be-recognized image, and the next to-be-recognized image can be continued to be recognized. By detecting whether the to-be-recognized image contains a face image, the to-be-recognized image without a face image is eliminated, and the efficiency of image detection is greatly improved.

[0037] Further, multiple face images can exist in a to-be-recognized image, and in this case, all the face images in the to-be-recognized image need to be recognized one by one. Therefore, the step of segmenting a face image from the to-be-recognized image and extracting a feature vector of the face image by using a pre-trained deep neural network to obtain a to-be-recognized feature vector specifically includes:

[0038] 1. Determine whether multiple face images exist in the to-be-recognized image.

[0039] Specifically, the detection result of the multi-task convolutional neural network is used to determine whether multiple face images exist in the to-be-identified image.

[0040] 2. If yes, at least one face image is segmented from the to-be-identified image, and a unique identifier is assigned to each face image.

[0041] Specifically, when multiple face images exist in the to-be-identified image, each face image is segmented, and a unique identifier is assigned to each face image for easy identification. The unique identifier can be implemented in the form of an image name of the to-be-identified image plus a serial number.

[0042] 3. The image features of each face image are extracted using a deep neural network to obtain a to-be-identified feature vector of each face image.

[0043] Specifically, after all face images in the to-be-identified image are segmented, feature extraction is performed on each face image.

[0044] It should be understood that when multiple face images of special face types exist in a to-be-identified image, all the special face types that exist are unified as the identification result of the to-be-identified image when the special face type identification result is output.

[0045] Step S103: The to-be-identified feature vector is compared with each label feature vector of a special face type to obtain a target special face type that meets a preset condition. The special face type is pre-set, and the label feature vector is calculated from image feature vectors extracted from multiple special face images, wherein the multiple special face images belong to multiple special face types.

[0046] It should be noted that the special face type refers to a pre-specified face type that needs to be monitored. In this embodiment, the special face type includes three types of real face, cartoon face, and sketch face.

[0047] Specifically, before using the face type identification method of the heterogeneous face of the embodiment to identify the face type, the special face type and the label feature vector corresponding to the special face type need to be prepared in advance. The label feature vector is calculated according to the image feature vectors of the special face images corresponding to each special face type.

[0048] Further, in order to ensure that the data is balanced enough, the number of special face images corresponding to each special face type is the same, and the special face images are all single face images.

[0049] Further, the label feature vector calculation process includes:

[0050] 1. Extract the image feature vector of each special face image using a pre-trained deep neural network.

[0051] Specifically, the 128-dimensional image feature vector of each special face image of each special face type is extracted by a pre-trained Facenet face recognition model.

[0052] 2. Calculate the mean feature vector and the standard deviation feature vector of each special face type according to all image feature vectors corresponding to each special face type, and the mean feature vector and the standard deviation feature vector constitute the label feature vector.

[0053] Specifically, after obtaining the image feature vector of each special face image, the image feature vectors of all special face images belonging to the same special face type are accumulated in dimension, and then divided by the number of all special face images belonging to the same special face type to obtain the mean of each dimension. The mean of all dimensions constitutes the mean feature vector. The standard deviation feature vector is calculated using the image feature vector and the mean feature vector. After obtaining the mean feature vector and the standard deviation feature vector, they are taken as the label feature vector of the special face type.

[0054] Further, the to-be-identified feature vector is compared with the label feature vector of each special face type to obtain the target special face type that meets the preset condition, including:

[0055] 1. Construct the feature value range of each dimension corresponding to each special face type using the feature values of each dimension of the mean feature vector and the standard deviation feature vector corresponding to each special face type.

[0056] Specifically, the feature value range of each dimension is: [feature value of mean feature vector-2*feature value of standard deviation feature vector, feature value of mean feature vector+2*feature value of standard deviation feature vector].

[0057] 2. Compare the feature values of each dimension of the to-be-identified feature vector with the feature value range of each dimension corresponding to each special face type to obtain the first special face type that meets the preset range condition, and the preset range condition includes that the feature values of each dimension of the to-be-identified feature vector fall within the feature value range of the corresponding dimension of the first special face type.

[0058] 3. Calculate the Euclidean distance between the image feature vector of each special face image in the first special face type and the to-be-identified feature vector.

[0059] 4. When there is at least one Euclidean distance that meets the preset distance condition, the first special face type is retained, otherwise the first special face type is deleted.

[0060] 5. Selecting the first special face type corresponding to the image feature vector with the minimum Euclidean distance as the target special face type.

[0061] Further, when there is no first special face type satisfying the preset range condition or no Euclidean distance satisfying the preset distance condition, the image to be identified is delivered to manual review.

[0062] Step S104: Outputting the target special face type as the face type of the image to be identified.

[0063] The face type recognition method for heterogeneous faces according to the embodiment of the application pre-constructs label feature vectors for representing special face types, extracts a feature vector to be identified in an image to be identified after obtaining the image to be identified, compares the feature vector to be identified with each label feature vector, confirms a target label feature vector closest to the feature vector to be identified and a target special face type corresponding to the target label feature vector, and outputs the target special face type as the face type to which the image to be identified belongs. The method does not need to rely on a deep neural network with a multi-layer structure and can realize recognition of special face images, has small calculation amount, fast response speed, and small system resource occupation while ensuring accuracy.

[0064] Figure 2 is a functional module schematic diagram of the face type recognition device for heterogeneous faces according to the embodiment of the application. As shown in the figure, the device 20 includes an acquisition module 21, an extraction module 22, a comparison module 23, and an output module 24. Figure 2

[0065] The acquisition module 21 is configured to acquire an image to be identified.

[0066] The extraction module 22 is configured to extract a feature vector to be identified from the image to be identified.

[0067] The comparison module 23 is configured to compare the feature vector to be identified with each label feature vector of a special face type, obtain a target special face type satisfying a preset condition, and pre-set the special face type. The label feature vector is calculated from image feature vectors extracted from a plurality of special face images, wherein the plurality of special face images belong to a plurality of special face types.

[0068] The output module 24 is configured to output the target special face type as the face type of the image to be identified.

[0069] ​Optionally, the comparison module 23 is further configured to pre-calculate the label feature vector, including: extracting an image feature vector of each special face image respectively by using a pre-trained deep neural network; calculating a mean feature vector and a standard deviation feature vector of each special face type respectively according to all image feature vectors corresponding to each special face type, and the mean feature vector and the standard deviation feature vector constitute the label feature vector.

[0070] Optionally, the comparison module 23 performs an operation of comparing the to-be-recognized feature vector with the label feature vector of each special face type respectively to obtain a target special face type satisfying a preset condition, and the operation specifically includes: constructing a feature value range of each dimension corresponding to each special face type by using the feature value of each dimension of the mean feature vector and the standard deviation feature vector corresponding to each special face type; comparing the feature value of each dimension of the to-be-recognized feature vector with the feature value range of each dimension corresponding to each special face type respectively to obtain a first special face type satisfying a preset range condition, and the preset range condition includes that the feature value of each dimension of the to-be-recognized feature vector falls into the feature value range of the corresponding dimension of the first special face type; calculating the Euclidean distance between the image feature vector of each special face image in the first special face type and the to-be-recognized feature vector; when there is at least one Euclidean distance satisfying a preset distance condition, retaining the first special face type, otherwise deleting the first special face type; selecting the first special face type corresponding to the image feature vector with the smallest Euclidean distance as the target special face type.

[0071] Optionally, the comparison module 23 is further configured to: when there is no first special face type satisfying the preset range condition or there is no Euclidean distance satisfying the preset distance condition, transport the to-be-recognized image to artificial audit.

[0072] Optionally, the extraction module 22 performs an operation of extracting the to-be-recognized feature vector from the to-be-recognized image, and the operation specifically includes: detecting whether there is a face image in the to-be-recognized image by using a pre-trained multi-task convolutional neural network; if there is, segmenting the face image from the to-be-recognized image, and then extracting the feature vector of the face image by using a pre-trained deep neural network to obtain the to-be-recognized feature vector; if there is not, skipping the current to-be-recognized image and continuing to recognize the next to-be-recognized image.

[0073] Optionally, the extraction module 22 performs the operation of segmenting the face image from the image to be identified, and then using a pre-trained deep neural network to extract the feature vector of the face image to obtain the feature vector to be identified. Specifically, this includes: determining whether there are multiple face images in the image to be identified; if so, segmenting at least one face image from the image to be identified and assigning a unique identifier to each face image; and using a deep neural network to extract the image features of each face image to obtain the feature vector to be identified for each face image.

[0074] Optionally, the number of special face images corresponding to each special face type is the same, and all special face images are single face images.

[0075] For further details regarding the implementation techniques of each module in the heterogeneous face type recognition device of the above embodiments, please refer to the description in the heterogeneous face type recognition method of the above embodiments, which will not be repeated here.

[0076] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0077] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. Figure 3 As shown, the computer device 30 includes a processor 31 and a memory 32 coupled to the processor 31. The memory 32 stores program instructions. When the program instructions are executed by the processor 31, the processor 31 performs the steps of the heterogeneous face type recognition method described in any of the above embodiments.

[0078] The processor 31 can also be referred to as a CPU (Central Processing Unit). The processor 31 may be an integrated circuit chip with signal processing capabilities. The processor 31 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0079] See Figure 4 , Figure 4A structural schematic diagram of the storage medium of the embodiment of the present application. The storage medium of the embodiment of the present application stores program instructions 41 capable of realizing all the methods described above, wherein the program instructions 41 can be stored in the storage medium in the form of a software product, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes, or a computer, a server, a mobile phone, a tablet computer, and the like.

[0080] In several embodiments provided in the present application, it should be understood that the disclosed computer device, apparatus and method can be implemented in other manners. For example, the described apparatus embodiments are merely schematic. The division of the units is merely a logical function division. There can be another division manner for the actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be electrical, mechanical or in other forms.

[0081] In addition, each function unit in the various embodiments of the present application can be integrated into a processing unit, or each unit can exist alone physically, or two or more units can be integrated into one unit. The integrated unit can be realized in the form of hardware, or in the form of a software functional unit. The above is merely an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent flow transformation using the content of the specification and drawings of the present application, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A method for recognizing face types in heterogeneous faces, characterized in that, include: Acquire the image to be recognized; The feature vector to be identified is extracted from the image to be identified; The feature vector to be identified is compared with the label feature vector of each special face type to obtain the target special face type that meets the preset conditions. The special face type is preset, and the label feature vector is calculated from the image feature vectors extracted from multiple special face images, wherein the multiple special face images belong to multiple special face types. The step of comparing the feature vector to be identified with the label feature vector of each specific face type to obtain the target specific face type that meets the preset conditions includes: The feature value range for each dimension of each face type is constructed by using the feature values ​​of the mean feature vector and standard deviation feature vector corresponding to each specific face type; The feature values ​​of each dimension of the feature vector to be identified are compared with the feature value range of each dimension corresponding to each special face type to obtain the first special face type that meets the preset range condition. The preset range condition includes that the feature values ​​of each dimension of the feature vector to be identified fall within the feature value range of the corresponding dimension of the first special face type. Calculate the Euclidean distance between the image feature vector of each special face image in the first special face type and the feature vector to be identified; If at least one Euclidean distance satisfies the preset distance condition, the first special face type is retained; otherwise, the first special face type is deleted. The first special face type corresponding to the image feature vector with the smallest Euclidean distance is selected as the target special face type; The target special face type is output as the face type of the image to be identified.

2. The face type recognition method for heterogeneous faces according to claim 1, characterized in that, The label feature vector calculation process includes: The image feature vector of each specific face image is extracted using a pre-trained deep neural network. The mean feature vector and standard deviation feature vector for each specific face type are calculated based on all image feature vectors corresponding to each specific face type. The mean feature vector and the standard deviation feature vector constitute the label feature vector.

3. The face type recognition method for heterogeneous faces according to claim 2, characterized in that, The method further includes: If there is no first special face type that meets the preset range conditions or no Euclidean distance that meets the preset distance conditions, the image to be identified will be sent for manual review.

4. The face type recognition method for heterogeneous faces according to claim 1, characterized in that, The feature vector to be identified is extracted from the image to be identified, including: The presence of a human face image in the image to be identified is detected using a pre-trained multi-task convolutional neural network. If it exists, the face image is obtained by segmenting the image to be identified, and then the feature vector of the face image is extracted using a pre-trained deep neural network to obtain the feature vector to be identified. If it does not exist, skip the current image to be identified and continue to identify the next image to be identified.

5. The face type recognition method for heterogeneous faces according to claim 4, characterized in that, The process of segmenting the face image from the image to be identified, and then extracting the feature vector of the face image using a pre-trained deep neural network to obtain the feature vector to be identified includes: Determine whether there are multiple face images in the image to be identified; If so, at least one face image is segmented from the image to be identified, and each face image is assigned a unique identifier; The deep neural network is used to extract the image features of each face image to obtain the feature vector to be identified for each face image.

6. The face type recognition method for heterogeneous faces according to claim 1, characterized in that, The number of special face images corresponding to each special face type is the same, and all of the special face images are single face images.

7. A face type recognition device for heterogeneous faces, characterized in that, include: The acquisition module is used to acquire the image to be recognized; The extraction module is used to extract the feature vector to be identified from the image to be identified; The comparison module is used to compare the feature vector to be identified with the label feature vector of each special face type to obtain the target special face type that meets the preset conditions. The special face type is preset, and the label feature vector is calculated from the image feature vectors extracted from multiple special face images, wherein the multiple special face images belong to multiple special face types. The step of comparing the feature vector to be identified with the label feature vector of each specific face type to obtain the target specific face type that meets the preset conditions includes: The feature value range for each dimension of each face type is constructed by using the feature values ​​of the mean feature vector and standard deviation feature vector corresponding to each specific face type; The feature values ​​of each dimension of the feature vector to be identified are compared with the feature value range of each dimension corresponding to each special face type to obtain the first special face type that meets the preset range condition. The preset range condition includes that the feature values ​​of each dimension of the feature vector to be identified fall within the feature value range of the corresponding dimension of the first special face type. Calculate the Euclidean distance between the image feature vector of each special face image in the first special face type and the feature vector to be identified; If at least one Euclidean distance satisfies the preset distance condition, the first special face type is retained; otherwise, the first special face type is deleted. The first special face type corresponding to the image feature vector with the smallest Euclidean distance is selected as the target special face type; The output module is used to output the target special face type as the face type of the image to be identified.

8. A computer device, characterized in that, The computer device includes a processor and a memory coupled to the processor, the memory storing program instructions that, when executed by the processor, cause the processor to perform the steps of the face type recognition method for heterogeneous faces as described in any one of claims 1-6.

9. A storage medium, characterized in that, The system stores program instructions capable of implementing the face type recognition method for heterogeneous faces as described in any one of claims 1-6.

Images