Subject face image recognition method, device, electronic device and storage medium

Abstract

The present disclosure relates to a subject face image recognition method, device, electronic device and storage medium, the method comprising performing face recognition on a target image including multiple faces to obtain multiple face images; performing multi-dimensional quality analysis on the multiple face images to obtain initial quality data; based on the preset constraint quality data corresponding to each analysis dimension, constraining the initial quality data under each analysis dimension to obtain the constrained quality result of each face image under each analysis dimension, the preset constraint quality data is used to balance the initial quality data under each analysis dimension, and the degree of influence on the quality of each face image; based on the constraint quality results of each face image under multiple analysis dimensions, generating target quality data for each face image; based on the target quality data, identifying the target subject face image from the multiple face images. The embodiments of the present disclosure can improve the breadth and accuracy of subject face recognition scenarios.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technology, and in particular to a subject face image recognition method, device, electronic device and storage medium. Background Art

[0002] In recent years, image processing technology based on face recognition has been widely used. For example, in the application scenario of identifying human actions in video images and using these actions to record and manipulate images, since there may be multiple people in the video image, it is often necessary to first combine face recognition technology to determine the face image of the main person from the many face images in the video image in order to accurately locate the action of the main person.

[0003] In the related art, the recognition of the subject facial image can mainly include: "the first face that appears", "the largest face", and other schemes. Among them, the "first face that appears" scheme has low recognition accuracy when dealing with the situation where there are multiple faces when the video starts. The "largest face" scheme usually sorts the faces based on the area they occupy on the screen, and selects the face image with the largest area as the subject facial image; but in the conference scene, since the camera is usually at a certain distance from the photographer, people will frequently pass between the camera and the photographer, and the people passing by will appear as larger faces in the video image, resulting in low recognition accuracy. The above-mentioned related art subject facial image recognition scheme cannot effectively cope with different scenarios, and has problems such as large scenario limitations and low subject facial image recognition accuracy. Summary of the invention

[0004] The present disclosure provides a subject face image recognition method, device, electronic device and storage medium to at least solve the problems of large scene limitations and low subject face image recognition accuracy in related technologies. The technical solution of the present disclosure is as follows:

[0005] According to a first aspect of an embodiment of the present disclosure, a subject face image recognition method is provided, comprising:

[0006] Performing face recognition on a target image including multiple faces to obtain multiple face images;

[0007] Performing multi-dimensional quality analysis on the multiple facial images to obtain initial quality data of each facial image in multiple analysis dimensions;

[0008] Based on the preset constraint quality data corresponding to each analysis dimension, the initial quality data under each analysis dimension is constrained to obtain the constrained quality result of each face image under each analysis dimension, wherein the preset constraint quality data is used to balance the initial quality data under each analysis dimension and the degree of influence on the quality of each face image;

[0009] generating target quality data for each face image based on the constrained quality results of each face image under the multiple analysis dimensions;

[0010] Based on the target quality data, a target subject face image is identified from the multiple face images.

[0011] Optionally, the identifying a target subject face image from the multiple face images based on the target quality data includes:

[0012] Determining, from the plurality of face images, a primary subject face image corresponding to the maximum target quality data;

[0013] The preliminarily selected subject face image is used as the target subject face image.

[0014] Optionally, the target image is a preset frame image in a preset video, the pre-selected subject face image is an image of the target face, and the method further includes:

[0015] Acquire subject face images corresponding to a first number of frame images before the preset frame image;

[0016] The using the pre-selected subject face image as the target subject face image comprises:

[0017] In a case where the subject facial images corresponding to the first number of frame images include images of the second number of target faces, the preliminarily selected subject facial images are used as the target subject facial images.

[0018] Optionally, the method further includes:

[0019] Acquire subject face images corresponding to a third number of frame images before the preset frame image;

[0020] In the case that the subject face images corresponding to the first number of frame images include the second number of target face images, using the preliminarily selected subject face images as the target subject face images comprises:

[0021] When the subject facial images corresponding to the first number of frame images include images of the second number of the target faces, and the subject facial images corresponding to the third number of frame images are all images of the target faces, the preliminarily selected subject facial images are used as the target subject facial images.

[0022] Optionally, the target image is a preset frame image in a preset video, and generating the target quality data of each face image based on the constraint quality results of each face image under the multiple analysis dimensions includes:

[0023] Determining intermediate quality data of each face image according to the constraint quality results of each face image under the multiple analysis dimensions;

[0024] Determine a target face image including a historical subject face from the plurality of face images, wherein the historical subject face is a face in a subject face image corresponding to a historical frame image of the target image;

[0025] Based on a preset scaling factor, scaling the intermediate quality data of the target face image to obtain target quality data of the target face image;

[0026] The intermediate quality data of other human face images are used as the target quality data of the other human face images, and the other human face images are human face images other than the target human face image among the multiple human face images.

[0027] Optionally, the target quality data of the target facial image includes reduced quality data or enlarged quality data; and the scaling process of the intermediate quality data of the target facial image based on a preset scale factor to obtain the target quality data of the target facial image includes:

[0028] Determining the preset scaling factor based on the target scene information;

[0029] When the preset scale factor is a negative number, the intermediate quality data of the target face image is reduced based on the preset scale factor to obtain the reduced quality data;

[0030] When the preset proportionality coefficient is a positive number, the intermediate quality data of the target face image is enlarged based on the preset proportionality coefficient to obtain the enlarged quality data.

[0031] Optionally, the preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes target constraint quality data;

[0032] The constraining process is performed on the initial quality data under each analysis dimension based on the preset constrained quality data corresponding to each analysis dimension to obtain the constrained quality result of each face image under each analysis dimension, including:

[0033] In a case where the multiple facial images include a first facial image, using an upper limit of quality data corresponding to a target analysis dimension as target constraint quality data of the first facial image under the target analysis dimension; the target analysis dimension is any analysis dimension among the multiple analysis dimensions, and the first facial image is a facial image among the multiple facial images, the initial quality data of which under the target analysis dimension is greater than the upper limit of quality data corresponding to the target analysis dimension;

[0034] In the case where the multiple facial images include a second facial image, the lower limit of the quality data corresponding to the target analysis dimension is used as the target constraint quality data of the second facial image under the target analysis dimension; the second facial image is a facial image among the multiple facial images, the initial quality data under the target analysis dimension of which is less than the lower limit of the quality data corresponding to the target analysis dimension;

[0035] In the case that the multiple facial images include a third facial image, the initial quality data of the third facial image under the target analysis dimension is used as the target constraint quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the multiple facial images, whose initial quality data under the target analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension.

[0036] Optionally, the preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes target constraint quality data;

[0037] The constraining process is performed on the initial quality data under each analysis dimension based on the preset constrained quality data corresponding to each analysis dimension to obtain the constrained quality result of each face image under each analysis dimension, including:

[0038] In a case where the multiple facial images include a first facial image, using an upper limit of quality data corresponding to a target analysis dimension as initial constrained quality data of the first facial image under the target analysis dimension; the target analysis dimension is any analysis dimension among the multiple analysis dimensions, and the first facial image is a facial image among the multiple facial images, the initial quality data of which under the target analysis dimension is greater than an upper limit of quality data corresponding to the target analysis dimension;

[0039] In the case where the multiple facial images include a second facial image, the lower limit of the quality data corresponding to the target analysis dimension is used as the initial constraint quality data of the second facial image under the target analysis dimension; the second facial image is a facial image among the multiple facial images whose initial quality data under the target analysis dimension is less than the lower limit of the quality data corresponding to the target analysis dimension;

[0040] In the case where the multiple facial images include a third facial image, initial quality data of the third facial image under the target analysis dimension is used as initial constrained quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the multiple facial images whose initial quality data under the target analysis dimension is greater than or equal to a lower limit of quality data corresponding to the target analysis dimension and less than an upper limit of quality data corresponding to the target analysis dimension;

[0041] Based on the initial constraint quality data, target constraint quality data of each face image in each analysis dimension is generated.

[0042] Optionally, the constraint quality result further includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality;

[0043] The constraining the initial quality data under each analysis dimension based on the preset constrained quality data corresponding to each analysis dimension to obtain the constrained quality result of each face image under each analysis dimension also includes:

[0044] Taking unqualified quality as a quality determination result of the second facial image;

[0045] The qualified quality is used as the quality judgment result of the first face image and the three face images.

[0046] Optionally, the constraint quality result includes target constraint quality data;

[0047] The generating target quality data of each face image based on the constraint quality results of each face image under the multiple analysis dimensions comprises:

[0048] Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of each face image under the multiple analysis dimensions to obtain the target quality data.

[0049] Optionally, the constraint quality result further includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality;

[0050] The generating target quality data of each face image based on the constraint quality results of each face image under the multiple analysis dimensions further comprises:

[0051] determining a key analysis dimension among the plurality of analysis dimensions;

[0052] Based on the quality determination results under the key analysis dimensions, the multiple face images are filtered to obtain face images to be screened;

[0053] The weighted summation of the target constraint quality data of each face image under the multiple analysis dimensions based on the preset quality weights corresponding to the multiple analysis dimensions to obtain the target quality data includes:

[0054] Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of the face image to be screened under the multiple analysis dimensions to obtain the target quality data;

[0055] The step of identifying a target subject face image from the plurality of face images based on the target quality data comprises:

[0056] Based on the target quality data, the target subject face image is identified from the face images to be screened.

[0057] According to a second aspect of an embodiment of the present disclosure, a subject face image recognition device is provided, comprising:

[0058] A face recognition module is configured to perform face recognition on a target image including multiple faces to obtain multiple face images;

[0059] A multi-dimensional quality analysis module is configured to perform multi-dimensional quality analysis on the plurality of face images to obtain initial quality data of each face image under a plurality of analysis dimensions;

[0060] a constraint processing module configured to execute, based on preset constraint quality data corresponding to each analysis dimension, constraint processing on the initial quality data under each analysis dimension to obtain the constraint quality result of each face image under each analysis dimension, wherein the preset constraint quality data is used to balance the initial quality data under each analysis dimension and the degree of influence on the quality of each face image;

[0061] A target quality data generating module is configured to generate target quality data for each face image based on the constraint quality results of each face image under the multiple analysis dimensions;

[0062] The target subject facial image recognition module is configured to identify a target subject facial image from the multiple facial images based on the target quality data.

[0063] Optionally, the target subject face image recognition module includes:

[0064] A preliminary subject face image recognition unit is configured to determine a preliminary subject face image corresponding to the maximum target quality data from the plurality of face images;

[0065] The target subject face image recognition unit is configured to execute the process of using the preliminary selected subject face image as the target subject face image.

[0066] Optionally, the target image is a preset frame image in a preset video, the pre-selected subject face image is an image of the target face, and the device further includes:

[0067] A first subject face image acquisition module is configured to execute acquisition of subject face images corresponding to a first number of frame images before the preset frame image;

[0068] The target subject face image recognition unit is further configured to perform:

[0069] In a case where the subject facial images corresponding to the first number of frame images include images of the second number of target faces, the preliminarily selected subject facial images are used as the target subject facial images.

[0070] Optionally, the device further comprises:

[0071] A second subject face image acquisition module is configured to execute acquisition of subject face images corresponding to a third number of frame images before the preset frame image;

[0072] The target subject face image recognition unit is further configured to perform:

[0073] When the subject facial images corresponding to the first number of frame images include images of the second number of the target faces, and the subject facial images corresponding to the third number of frame images are all images of the target faces, the preliminarily selected subject facial images are used as the target subject facial images.

[0074] Optionally, the target image is a preset frame image in a preset video, and the target quality data generating module includes:

[0075] an intermediate quality data determining unit, configured to determine the intermediate quality data of each face image according to the constraint quality results of each face image under the multiple analysis dimensions;

[0076] A target facial image determining unit is configured to determine a target facial image including a historical subject face from the plurality of facial images, wherein the historical subject face is a face in a subject facial image corresponding to a historical frame image of the target image;

[0077] A scaling processing unit is configured to perform scaling processing on the intermediate quality data of the target face image based on a preset scaling coefficient to obtain target quality data of the target face image;

[0078] The target quality data determination unit is configured to use the intermediate quality data of other human face images as the target quality data of the other human face images, wherein the other human face images are human face images other than the target human face image among the multiple human face images.

[0079] Optionally, the target quality data of the target facial image includes reduced quality data or enlarged quality data; and the scaling processing unit includes:

[0080] A preset scale factor determination unit is configured to determine the preset scale factor based on target scene information;

[0081] a reduction processing unit configured to perform reduction processing on the intermediate quality data of the target face image based on the preset scale factor when the preset scale factor is a negative number, so as to obtain the reduced quality data;

[0082] The enlargement processing unit is configured to perform enlargement processing on the intermediate quality data of the target face image based on the preset scale factor when the preset scale factor is a positive number, so as to obtain the enlarged quality data.

[0083] Optionally, the preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes target constraint quality data;

[0084] The constraint processing module comprises:

[0085] The first target constraint quality data determination unit is configured to, when the plurality of face images include a first face image, use an upper limit of quality data corresponding to a target analysis dimension as target constraint quality data of the first face image under the target analysis dimension; the target analysis dimension is any analysis dimension among the plurality of analysis dimensions, and the first face image is a face image among the plurality of face images, the initial quality data under the target analysis dimension being greater than the upper limit of quality data corresponding to the target analysis dimension;

[0086] The second target constraint quality data determination unit is configured to use, when the plurality of face images include a second face image, a lower limit of quality data corresponding to the target analysis dimension as the target constraint quality data of the second face image under the target analysis dimension; the second face image is a face image among the plurality of face images, the initial quality data under the target analysis dimension of which is less than the lower limit of quality data corresponding to the target analysis dimension;

[0087] The third target constraint quality data determination unit is configured to execute, when the multiple facial images include a third facial image, using the initial quality data of the third facial image under the target analysis dimension as the target constraint quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the multiple facial images, whose initial quality data under the target analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension.

[0088] Optionally, the preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes target constraint quality data;

[0089] The constraint processing module comprises:

[0090] The first initial constrained quality data determination unit is configured to, when the plurality of facial images include a first facial image, use an upper limit of quality data corresponding to a target analysis dimension as initial constrained quality data of the first facial image under the target analysis dimension; the target analysis dimension is any analysis dimension among the plurality of analysis dimensions, and the first facial image is a facial image among the plurality of facial images, the initial quality data under the target analysis dimension being greater than an upper limit of quality data corresponding to the target analysis dimension;

[0091] The second initial constrained quality data determining unit is configured to, when the plurality of facial images include a second facial image, use the lower limit of the quality data corresponding to the target analysis dimension as the initial constrained quality data of the second facial image under the target analysis dimension; the second facial image is a facial image among the plurality of facial images whose initial quality data under the target analysis dimension is less than the lower limit of the quality data corresponding to the target analysis dimension;

[0092] A third initial constrained quality data determination unit is configured to, when the plurality of facial images include a third facial image, use the initial quality data of the third facial image under the target analysis dimension as the initial constrained quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the plurality of facial images, the initial quality data of which under the target analysis dimension is greater than or equal to a lower limit of the quality data corresponding to the target analysis dimension and less than an upper limit of the quality data corresponding to the target analysis dimension;

[0093] The target constraint quality data generating unit is configured to generate the target constraint quality data of each face image in each analysis dimension based on the initial constraint quality data.

[0094] Optionally, the constraint quality result further includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality;

[0095] The constraint processing module also includes:

[0096] A first quality determination result determining unit is configured to perform unqualified quality determination as a quality determination result of the second facial image;

[0097] The second quality determination result determining unit is configured to use qualified quality as the quality determination result of the first face image and the three-person face images.

[0098] Optionally, the constraint quality result includes target constraint quality data;

[0099] The target quality data generation module includes:

[0100] The weighted summation unit is configured to perform weighted summation of the target constraint quality data of each face image under the multiple analysis dimensions based on preset quality weights corresponding to the multiple analysis dimensions to obtain the target quality data.

[0101] Optionally, the constraint quality result further includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality;

[0102] The target quality data generation module also includes:

[0103] A key analysis dimension determination unit, configured to determine a key analysis dimension among the plurality of analysis dimensions;

[0104] An image filtering unit is configured to perform filtering on the plurality of face images based on the quality determination result under the key analysis dimension to obtain a face image to be screened;

[0105] The weighted summation unit is further configured to perform:

[0106] Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of the face image to be screened under the multiple analysis dimensions to obtain the target quality data;

[0107] The target subject facial image recognition module is further configured to identify the target subject facial image from the facial images to be screened based on the target quality data.

[0108] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement a method as described in any one of the above-mentioned first aspects.

[0109] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute any method described in the first aspect of the embodiment of the present disclosure.

[0110] According to a fifth aspect of an embodiment of the present disclosure, there is provided a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any of the methods described in the first aspect of the embodiment of the present disclosure.

[0111] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0112] In the process of subject face recognition, by performing multi-dimensional quality analysis on multiple face images in the target image, it is possible to perform quality analysis on face images from multiple dimensions, effectively meet the needs of different scenarios, and improve the breadth of subject face recognition scenarios. In addition, based on the preset constraint quality data corresponding to each analysis dimension, the initial quality data under each analysis dimension is constrained, which can effectively balance the quality impact of the quality data of multiple analysis dimensions on the face image; then, based on the constraint quality results of each face image under multiple analysis dimensions, the target quality data of each face image is generated, and the target subject face image is identified from multiple face images, which can greatly improve the accuracy of subject face recognition.

[0113] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0114] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0115] Figure 1 is a flow chart of a subject face image recognition method according to an exemplary embodiment;

[0116] Figure 2 A flowchart showing a method of performing constraint processing on initial quality data under each analysis dimension based on preset constraint quality data corresponding to each analysis dimension to obtain a constraint quality result of each face image under each analysis dimension according to an exemplary embodiment;

[0117] Figure 3 is a block diagram of a subject face image recognition device according to an exemplary embodiment;

[0118] Figure 4 The present invention is a block diagram of an electronic device for subject face image recognition according to an exemplary embodiment. DETAILED DESCRIPTION

[0119] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0120] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0121] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0122] See also Figure 1 , Figure 1 is a flow chart of a subject face image recognition method according to an exemplary embodiment. Figure 1As shown, the subject face image recognition method is used in devices such as terminals, servers, edge computing nodes, and can also be used in software running on the device, such as applications, etc. Specifically, it can include the following steps.

[0123] In step S101, face recognition is performed on a target image including multiple faces to obtain multiple face images.

[0124] In the embodiments of this specification, the target image may be an image including multiple faces. In a specific embodiment, the target image may be a single captured image, or a preset frame image in a preset video. Optionally, the preset video may be a pre-recorded video, or a real-time recorded video; optionally, the preset frame image may be a frame image determined in combination with an actual application scenario, which requires subject face image recognition. In a specific embodiment, taking the preset video as a real-time recorded video at an event site as an example, the target image may be each current frame image (preset frame image) in the preset video.

[0125] In a specific embodiment, face recognition can be performed on the target image based on a face recognition network to obtain multiple face images included in the target image. Optionally, the output of the face recognition network may include the number of faces, the location information of each face, the facial key point data of each face, and the identification information of each face. Accordingly, the location information of each face and the facial key point data of each face can be combined to extract multiple face images from the target image. Specifically, the identification information of each face can be used to distinguish the face images of different people appearing in the video.

[0126] In step S103, a multi-dimensional quality analysis is performed on the multiple face images to obtain initial quality data of each face image in multiple analysis dimensions.

[0127] In a specific embodiment, the initial quality data of each face image in any analysis dimension may be an initial parameter that can measure the quality of each face image in the analysis dimension. Optionally, the better the quality of the face image in the analysis dimension, the larger the initial quality data.

[0128] In a specific embodiment, the multiple analysis dimensions may include but are not limited to at least two of a face size analysis dimension, a face orientation analysis dimension, a face position analysis dimension, and a gaze tracking analysis dimension.

[0129] In a specific embodiment, taking the quality analysis of multiple facial images in the facial size analysis dimension as an example, the first size information of the target image and the second size information of each facial image can be obtained; the ratio data between the second size information and the first size information is used as the initial quality data of each facial image in the facial size analysis dimension. Specifically, the first size information can be the size information of the screen used to display the target image, and correspondingly, the second size information can be the size information of the area occupied by each facial image on the screen. Optionally, the size information corresponding to the minimum circumscribed rectangular frame of each face can be used as the second size information, thereby simplifying the calculation of facial size information and improving the running speed and performance.

[0130] In a specific embodiment, taking the quality analysis of multiple facial images in the facial orientation analysis dimension as an example, the facial posture information and preset reference posture information of each facial image can be obtained; based on the facial posture information and reference posture information of each facial image, the initial quality data of each facial image in the facial orientation analysis dimension is determined.

[0131] In a specific embodiment, the reference posture information may be the facial posture information when a person stands at a certain position and looks directly at the target screen. The target screen may be the screen corresponding to the display device that displays the target image. In a specific embodiment, the facial posture information may be facial postures in three directions. Specifically, the facial postures in three directions may include a first tilt amplitude of the face in a first direction, a second tilt amplitude in a second direction, and a third tilt amplitude in a third direction. Specifically, the first direction may be the y-axis direction; accordingly, the first tilt amplitude may be the rotation angle of the face relative to the y-axis (left and right); the second direction may be the x-axis direction, and accordingly, the second tilt amplitude may be the rotation angle of the face relative to the x-axis (up and down); the third direction may be the z-axis direction, and accordingly, the third tilt amplitude may be the rotation angle of the face relative to the z-axis (vertical).

[0132] In practical applications, when capturing images, the position of the camera device is relatively fixed, and the reference posture information can be determined based on the relative position layout between the camera device and a display for displaying the target image.

[0133] In a specific embodiment, the facial posture information of each facial image can be determined by combining the relative position relationship between facial key points in each facial image, or by performing facial posture recognition on each facial image based on a preset facial posture recognition network. Specifically, the corresponding facial posture information determination method can be selected in combination with actual application requirements.

[0134] In an optional embodiment, the initial quality data of each face image under the facial orientation analysis dimension may be inversely proportional to the difference between the facial posture information of each face image and the reference posture information, that is, the greater the difference between the facial posture information of each face image and the reference posture information (the more biased the facial orientation is), the worse the image quality is, and the smaller the initial quality data is. In an optional embodiment, the initial quality data of the face image under the facial orientation analysis dimension may include the initial quality data corresponding to the above three directions.

[0135] In a specific embodiment, the initial quality data corresponding to any of the above directions can be calculated by combining the following formula: rscore = 1-math.abs(rB). Among them, rscore is the initial quality data corresponding to a certain direction; r is the facial posture of the face image in this direction; B is the reference facial posture in this direction in the reference posture information; math.abs() means taking the absolute value of the value in the brackets.

[0136] In an optional embodiment, taking the quality analysis of multiple face images in the face position analysis dimension as an example, the center position information (pixel coordinates) of the target image and the center position information (pixel coordinates) of each face image can be determined; the distance between the center position information of each face image and the center position information of the target image is calculated, and accordingly, the initial quality data of each face image is determined based on the distance. Optionally, the distance is inversely proportional to the initial quality data; specifically, the initial quality data of each face image determined based on the distance can be set in combination with the actual application, for example, the distance is used as the initial quality data, etc.

[0137] In an optional embodiment, taking the quality analysis of multiple facial images in the eye tracking analysis dimension as an example, the eye line angle corresponding to each facial image can be determined based on the relative position of the pupil and the surrounding area of ​​the eye in each facial image; and the reference eye line angle corresponding to the face facing the target screen can be determined; then, based on the eye line angle corresponding to each facial image and the reference eye line angle, the initial quality data of each facial image in the eye tracking analysis dimension is determined. Specifically, the specific details of determining the initial quality data of each facial image in the eye tracking analysis dimension based on the eye line angle corresponding to each facial image and the reference eye line angle can be referred to the above-mentioned determination of the initial quality data of each facial image in the facial orientation analysis dimension based on the facial posture information and reference posture information of each facial image.

[0138] In step S105, based on the preset constrained quality data corresponding to each analysis dimension, the initial quality data under each analysis dimension is constrained to obtain the constrained quality result of each face image under each analysis dimension.

[0139] In a specific embodiment, the preset constraint quality data can be used to balance the initial quality data under each analysis dimension and the degree of impact on the quality of each face image. In an optional embodiment, the preset constraint quality data can include a quality data upper limit and a quality data lower limit.

[0140] In an optional embodiment, the above-mentioned constraint quality results may include target constraint quality data. Specifically, the target constraint quality data under a certain analysis dimension may be a parameter that can measure the quality of the face image under the analysis dimension after balancing the initial quality data under the analysis dimension and the degree of impact on the quality of the face image.

[0141] In an optional embodiment, when the constrained quality result includes target constrained quality data, the initial quality data under each analysis dimension is constrained based on the preset constrained quality data corresponding to each analysis dimension, and the constrained quality result of each face image under each analysis dimension may include:

[0142] In a case where the plurality of facial images include the first facial image, using the upper limit of the quality data corresponding to the target analysis dimension as the target constraint quality data of the first facial image under the target analysis dimension;

[0143] In the case where the plurality of facial images include the second facial image, using the lower limit of the quality data corresponding to the target analysis dimension as the target constraint quality data of the second facial image under the target analysis dimension;

[0144] In the case where the plurality of facial images may include a third facial image, using initial quality data of the third facial image in the target analysis dimension as target constrained quality data of the third facial image in the target analysis dimension;

[0145] In a specific embodiment, the target analysis dimension may be any one of multiple analysis dimensions.

[0146] In an optional embodiment, the initial quality data of each facial image under the target analysis dimension can be compared with the numerical values ​​of the upper and lower limits of the quality data corresponding to the target analysis dimension. If the initial quality data of a facial image under the target analysis dimension is greater than the upper limit of the quality data corresponding to the target analysis dimension, the facial image can be used as the first facial image. Accordingly, the first facial image can be a facial image among multiple facial images whose initial quality data under the target analysis dimension is greater than the upper limit of the quality data corresponding to the target analysis dimension.

[0147] Optionally, when the initial quality data of a certain facial image under the target analysis dimension is less than the lower limit of the quality data corresponding to the target analysis dimension, the facial image can be used as the second facial image. Accordingly, the second facial image can be a facial image among multiple facial images whose initial quality data under the target analysis dimension is less than the lower limit of the quality data corresponding to the target analysis dimension.

[0148] Optionally, in the case of a facial image whose initial quality data under the target analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension, the facial image can be used as the third facial image. Accordingly, the third facial image is a facial image among multiple facial images whose initial quality data under the target analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension.

[0149] In a specific embodiment, taking the initial quality data of the above-mentioned facial image in the facial position analysis dimension as an example, if the initial quality data of a facial image in the facial position analysis dimension is greater than the upper limit of the quality data corresponding to the facial position analysis dimension, the upper limit of the quality data corresponding to the facial position analysis dimension can be used as the target constraint quality data of the facial image in the facial position analysis dimension. If the initial quality data of a facial image in the facial position analysis dimension is less than the lower limit of the quality data corresponding to the facial position analysis dimension, the lower limit of the quality data corresponding to the facial position analysis dimension can be used as the target constraint quality data of the facial image in the facial position analysis dimension. If the initial quality data of a facial image in the facial position analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the facial position analysis dimension, and is less than or equal to the upper limit of the quality data corresponding to the facial position analysis dimension, the initial quality data of the facial image in the facial position analysis dimension can be used as the target constraint quality data of the facial image in the facial position analysis dimension.

[0150] In the above embodiment, the quality data upper limit and the quality data lower limit corresponding to each analysis dimension are combined, and the quality data of multiple facial images in each analysis dimension are constrained within the range corresponding to the quality data upper limit and the quality data lower limit. This can effectively balance the impact of the quality data of multiple analysis dimensions on the quality of facial images, thereby improving the accuracy of the target quality data for measuring the quality of facial images.

[0151] In another optional embodiment, when the constraint quality result includes target constraint quality data, such as Figure 2 As shown, based on the preset constraint quality data corresponding to each analysis dimension, the initial quality data under each analysis dimension is constrained, and the constraint quality result of each face image under each analysis dimension may include:

[0152] In step S201, when the plurality of face images include a first face image, the upper limit of the quality data corresponding to the target analysis dimension is used as the initial constrained quality data of the first face image under the target analysis dimension;

[0153] In step S203, when the plurality of facial images include the second facial image, the lower limit of the quality data corresponding to the target analysis dimension is used as the initial constrained quality data of the second facial image under the target analysis dimension;

[0154] In step S205, when the plurality of facial images include a third facial image, the initial quality data of the third facial image in the target analysis dimension is used as the initial constrained quality data of the third facial image in the target analysis dimension;

[0155] In step S207, target constraint quality data for each face image in each analysis dimension is generated based on the initial constraint quality data.

[0156] In a specific embodiment, taking the initial quality data of the above-mentioned facial image under the facial orientation analysis dimension as an example, since the initial quality data of the facial image under the facial orientation analysis dimension includes initial quality data corresponding to three directions, correspondingly, the upper limit and lower limit of quality data corresponding to the facial orientation analysis dimension may include the upper limit and lower limit of quality data corresponding to each of the three directions.

[0157] In a specific embodiment, if the initial quality data corresponding to a certain direction of a facial image is greater than the upper limit of the quality data corresponding to the direction, the upper limit of the quality data corresponding to the direction may be used as the initial constrained quality data corresponding to the facial image in the direction. If the initial quality data corresponding to a certain direction of a facial image is less than the lower limit of the quality data corresponding to the direction, the lower limit of the quality data corresponding to the direction may be used as the initial constrained quality data corresponding to the facial image in the direction. If the initial quality data corresponding to a certain direction of a facial image is greater than or equal to the lower limit of the quality data corresponding to the direction, and less than or equal to the upper limit of the quality data corresponding to the direction, the initial quality data corresponding to the facial image in the direction may be used as the initial constrained quality data corresponding to the facial image in the direction.

[0158] Furthermore, the above-mentioned generation of target constraint quality data for each face image in each analysis dimension based on the initial constraint quality data may include: based on the preset weights corresponding to the three directions, weighting the initial constraint quality data corresponding to the face image in the three directions to obtain the target constraint quality data of the face image in the face orientation analysis dimension. Specifically, the preset weights may be configured in combination with the application scenario requirements. For example, in a certain application scenario, the corresponding direction needs to be controlled by shaking the face left and right. Accordingly, even if the rotation angle of the face relative to the first direction (left and right) is large, it should not affect the determination of the face as the main face. Accordingly, the preset weight corresponding to the first direction may be lower.

[0159] In another specific embodiment, taking the initial quality data of the above-mentioned facial image in the facial size analysis dimension as an example, if the initial quality data of a facial image in the facial size analysis dimension is greater than the upper limit of the quality data corresponding to the facial size analysis dimension, the upper limit of the quality data corresponding to the facial size analysis dimension can be used as the initial constrained quality data of the facial image in the facial size analysis dimension. If the initial quality data of a facial image in the facial size analysis dimension is less than the lower limit of the quality data corresponding to the facial size analysis dimension, the lower limit of the quality data corresponding to the facial size analysis dimension can be used as the initial constrained quality data of the facial image in the facial size analysis dimension. If the initial quality data of a facial image in the facial size analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the facial size analysis dimension, and is less than or equal to the upper limit of the quality data corresponding to the facial size analysis dimension, the initial quality data of the facial image in the facial size analysis dimension can be used as the initial constrained quality data of the facial image in the facial size analysis dimension.

[0160] Furthermore, based on the initial constraint quality data, the target constraint quality data of the face image in the face size analysis dimension can be generated in combination with the following formula: FSizeScore = (R-RMin) / (RMax-RMin). Among them, FSizeScore is the target constraint quality data of the face image in the face size analysis dimension, R is the initial constraint quality data of the face image in the face size analysis dimension, RMin is the lower limit of the quality data corresponding to the face size analysis dimension, and RMax is the upper limit of the quality data corresponding to the face size analysis dimension.

[0161] In the above embodiment, the initial quality data of multiple facial images in each analysis dimension are constrained in combination with the upper and lower limits of the quality data corresponding to each analysis dimension, and the target constrained quality data of each facial image in each analysis dimension is generated in combination with the initial constrained quality data. This can effectively balance the impact of the quality data of multiple analysis dimensions on the quality of the facial image, thereby improving the accuracy of the target quality data for measuring the quality of facial images.

[0162] In an optional embodiment, the above-mentioned constraint quality result may also include a quality judgment result. Specifically, the quality judgment result may include qualified quality and / or unqualified quality; optionally, the quality judgment result under a certain analysis dimension may be a condition for whether a facial image can participate in the subject facial image recognition.

[0163] Accordingly, the above-mentioned constrained processing is performed on the initial quality data under each analysis dimension based on the preset constrained quality data corresponding to each analysis dimension to obtain the constrained quality result of each face image under each analysis dimension, and may also include:

[0164] Taking unqualified quality as a quality determination result of the second face image;

[0165] The qualified quality is taken as the quality judgment result of the first face image and the three face images.

[0166] In a specific embodiment, when it is determined that the initial quality data of a facial image under the target analysis dimension is less than the lower limit of the quality data corresponding to the target analysis dimension, the lower limit of the quality data corresponding to the target analysis dimension can be used as the target constraint quality data of the facial image (second facial image) under the target analysis dimension, and unqualified quality can also be used as the quality judgment result of the facial image (second facial image).

[0167] In a specific embodiment, when it is determined that the initial quality data of a facial image under the target analysis dimension is greater than the upper limit of the quality data corresponding to the target analysis dimension, or is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension, the qualified quality can be used as the quality judgment result of the facial image (the first facial image or the third facial image).

[0168] In the above embodiment, the initial quality data under a certain analysis dimension is compared with the lower limit of the quality data corresponding to the analysis dimension to generate a quality judgment result that characterizes whether the image quality is sufficient to participate in subject face recognition. This can facilitate the subsequent filtering of facial images that are too poor to be subject face images, reduce the amount of data processing during subject face recognition, and improve the efficiency and accuracy of subject face recognition.

[0169] In step S107, target quality data of each face image is generated based on the constraint quality results of each face image under multiple analysis dimensions.

[0170] In an optional embodiment, the generation of target quality data for each face image based on the constraint quality results of each face image under multiple analysis dimensions may include:

[0171] Based on the preset quality weights corresponding to the multiple analysis dimensions, the target constraint quality data of each face image under the multiple analysis dimensions are weighted and summed to obtain the target quality data.

[0172] In a specific embodiment, the preset quality weight corresponding to each analysis dimension can be set in combination with the importance of the analysis dimension to the quality of the face image in actual applications. Specifically, the importance is proportional to the preset quality weight.

[0173] In the above embodiment, by combining the preset quality weights corresponding to multiple analysis dimensions and performing weighted summation on the target constraint quality data of each face image under multiple analysis dimensions, the target quality data that can characterize the quality of the face image can be obtained more accurately.

[0174] In an optional embodiment, when the target image is a preset frame image in a preset video, the above-mentioned generation of target quality data for each facial image based on the constraint quality results of each facial image under the multiple analysis dimensions may include the following steps: determining the intermediate quality data of each facial image according to the constraint quality results of each facial image under the multiple analysis dimensions; determining a target facial image including the face of a historical subject from the multiple facial images; scaling the intermediate quality data of the target facial image based on a preset scale coefficient to obtain the target quality data of the target facial image; and using the intermediate quality data of other facial images as the target quality data of the other facial images.

[0175] In a specific embodiment, determining the intermediate quality data of each facial image based on the constraint quality results of each facial image under the multiple analysis dimensions may include: based on preset quality weights corresponding to the multiple analysis dimensions, performing weighted summation of the target constraint quality data of each facial image under the multiple analysis dimensions to obtain the intermediate quality data.

[0176] In a specific embodiment, the preset quality weight corresponding to each analysis dimension can be set in combination with the importance of the analysis dimension to the quality of the face image in actual applications. Specifically, the importance is proportional to the preset quality weight.

[0177] In a specific embodiment, the historical subject face may be a face in a subject face image corresponding to a historical frame image of the target image; the historical frame image may include at least one frame image preceding the target image.

[0178] In an optional embodiment, the target quality data of the target facial image includes reduced quality data or enlarged quality data; the scaling process based on a preset scale factor on the intermediate quality data of the target facial image to obtain the target quality data of the target facial image may include:

[0179] Determining the preset scaling factor based on the target scene information;

[0180] When the preset scale factor is a negative number, the intermediate quality data of the target face image is reduced based on the preset scale factor to obtain the reduced quality data;

[0181] When the preset proportionality coefficient is a positive number, the intermediate quality data of the target face image is enlarged based on the preset proportionality coefficient to obtain the enlarged quality data.

[0182] In a specific embodiment, the preset scale factor can be preset in combination with the actual target scene information. Optionally, a mapping relationship between different target scene information and preset scale factors can be pre-established, and then the preset scale factor corresponding to a certain target scene information can be determined based on the mapping relationship. Specifically, the target scene information can be pre-set scene identification information.

[0183] In actual applications, two people often stand side by side. Accordingly, the target quality data of the facial images of the two people are very similar, which easily causes the recognized target subject facial image to switch back and forth between the two people. In some scenarios where operations need to be performed in conjunction with the actions of a certain person, that is, in scenarios where the target subject facial image is fixed to the facial image of a certain person, after obtaining the target quality data of the facial image, if a certain facial image is the subject facial image corresponding to the historical frame image, the intermediate quality data of the facial image can be optionally amplified based on a preset scale factor to avoid the target subject facial image switching back and forth between the two people and improve the accuracy of the subject facial image.

[0184] In an optional embodiment, in a scenario where the target subject facial image is fixed to a facial image of a certain person, the preset scaling factor can be a positive number, such as 20%. Accordingly, the target quality data of the target facial image can be multiplied by 1.2 to serve as the target quality data after scaling the target facial image.

[0185] In another specific embodiment, in some scenarios where the target subject face image needs to switch back and forth among multiple faces, the intermediate quality data of the face image can be reduced based on the preset scale factor to better ensure that the target subject face image switches back and forth among different people and improve the accuracy of the subject face image. Optionally, in the scenario where the target subject face image needs to switch back and forth among multiple faces, the preset scale factor can be set to a negative number, such as -20%, and accordingly, the target quality data of the target face image can be multiplied by 0.8 as the target quality data after scaling of the target face image.

[0186] In a specific embodiment, the other facial images mentioned above may be facial images other than the target facial image among the multiple facial images.

[0187] In the above embodiment, by setting the preset proportional coefficient in combination with different scenarios, it is possible to scale the target quality data of the target facial image according to the scenario requirements, thereby better responding to the requirements of subject facial recognition in different scenarios and improving the accuracy of the subject facial images identified in different scenarios.

[0188] In step S109, a target subject face image is identified from a plurality of face images based on the target quality data.

[0189] In a specific embodiment, the target subject facial image may be a facial image of a target user off-site corresponding to the target image. In an optional embodiment, the above-mentioned identifying the target subject facial image from multiple facial images based on the target quality data may include:

[0190] Determine, from among the multiple face images, a primary subject face image corresponding to the maximum target quality data;

[0191] The preliminary subject face image is used as the target subject face image.

[0192] In a specific embodiment, the subject face image corresponding to the maximum target quality data among the multiple face images may be used as the target subject face image.

[0193] In the above embodiment, the subject face image corresponding to the maximum target quality data is used as the target subject face image, which can comprehensively consider the quality of multiple analysis dimensions, effectively respond to the subject face recognition needs in various scenarios, and better improve the accuracy of the recognized subject face image.

[0194] In an optional embodiment, when the target image is a preset frame image in a preset video, the method may further include:

[0195] Acquire subject face images corresponding to a first number of frame images before a preset frame image;

[0196] Using the preliminary subject face image as the target subject face image includes:

[0197] In the case that the subject facial images corresponding to the first number of frame images include images of the second number of target faces, the preliminarily selected subject facial images are used as target subject facial images.

[0198] In a specific embodiment, the subject face images are preliminarily selected as the target face images; the first number and the second number can be preset in combination with actual applications, and the first number is greater than or equal to the second number.

[0199] In the above embodiment, in the scenario of identifying the subject face in the video, after the preliminary subject face image is determined in combination with the target quality data, if the subject face images corresponding to the first number of frame images before the preset frame image include the image of the target face in the second number of preliminary subject face images, the preliminary subject face image is used as the target subject face image, which can effectively combine the stability of the subject face images corresponding to multiple frame images to better improve the accuracy of the determined target subject face image.

[0200] In an optional embodiment, the above method may further include:

[0201] Acquire subject face images corresponding to a third number of frame images before a preset frame image;

[0202] In a case where the subject facial images corresponding to the first number of frame images include images of a second number of target faces, using the preliminarily selected subject facial images as target subject facial images may include:

[0203] When the subject facial images corresponding to the first number of frame images include images of the second number of target faces, and the subject facial images corresponding to the third number of frame images are all images of the target faces, the preliminarily selected subject facial images are used as target subject facial images.

[0204] In a specific embodiment, the third number may be preset in combination with actual applications, and optionally, the second number is greater than or equal to the third number.

[0205] In the above embodiment, in the scenario of identifying the subject face in the video, after the preliminary subject face image is determined in combination with the target quality data, the subject face images corresponding to the first number of frame images before the preset frame image include the images of the target face in the second number of preliminary subject face images, and the subject face images corresponding to the third number of frame images before the preset frame image are all images of the target face, then the preliminary subject face image is used as the target subject face image, which can effectively combine the stability of the subject face images corresponding to multiple consecutive frame images, and better improve the accuracy of the determined target subject face image.

[0206] In an optional embodiment, when the above-mentioned constrained quality result may further include a quality determination result; the above-mentioned generating target quality data for each face image based on the constrained quality results of each face image under multiple analysis dimensions may further include:

[0207] Identify key analysis dimensions among multiple analysis dimensions;

[0208] Based on the quality determination results under the key analysis dimensions, multiple face images are filtered to obtain face images to be screened.

[0209] Accordingly, the target quality data obtained by weighted summing the target constraint quality data of each face image under multiple analysis dimensions based on the preset quality weights corresponding to the multiple analysis dimensions may include:

[0210] Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of the face image to be screened under the multiple analysis dimensions to obtain the target quality data;

[0211] Based on the target quality data, the target subject face image is identified from multiple face images, including:

[0212] Based on the target quality data, the target subject face image is identified from the face images to be screened.

[0213] In an optional embodiment, when the above-mentioned constrained quality result may further include a quality determination result; the above-mentioned determining the intermediate quality data of each face image according to the constrained quality result of each face image under the multiple analysis dimensions further includes:

[0214] Identify key analysis dimensions among multiple analysis dimensions;

[0215] Based on the quality determination results under the key analysis dimensions, multiple face images are filtered to obtain face images to be screened.

[0216] Accordingly, the preset quality weights corresponding to the multiple analysis dimensions are used to perform weighted summation on the target constraint quality data of each face image under the multiple analysis dimensions, and the intermediate quality data obtained may include:

[0217] Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of the face image to be screened under the multiple analysis dimensions to obtain intermediate quality data;

[0218] Based on the target quality data, the target subject face image is identified from multiple face images, including:

[0219] Based on the target quality data, the target subject face image is identified from the face images to be screened.

[0220] In a specific embodiment, the key analysis dimension may be an analysis dimension that has a greater impact on the application scenario of the face image, and may be set in advance in combination with the actual application scenario requirements. For example, in a game scenario that requires forward and backward movement, if the face is too small, the movement information cannot be accurately identified, and accordingly, the face size analysis dimension may be a key analysis dimension.

[0221] In an optional embodiment, the key analysis dimension may include one or more analysis dimensions. Specifically, among the multiple facial images, facial images with unqualified quality determination results under any key analysis dimension may be filtered out to obtain facial images to be screened.

[0222] In the above embodiment, when the quality judgment result of a certain facial image under the key analysis dimension is unqualified, the facial image is directly filtered out, which can effectively reduce the data processing amount in the subject face recognition process and improve the subject face recognition efficiency and accuracy.

[0223] It can be seen from the technical solutions provided by the embodiments of this specification that in the subject face recognition process, this specification can realize the quality analysis of face images from multiple dimensions by performing multi-dimensional quality analysis on multiple face images in the target image, effectively respond to the needs of different scenarios, and improve the extensiveness of subject face recognition scenarios. In addition, based on the preset constraint quality data corresponding to each analysis dimension, the initial quality data under each analysis dimension is constrained, which can effectively balance the quality impact of the quality data of multiple analysis dimensions on the face image; then, based on the constraint quality results of each face image under multiple analysis dimensions, the target quality data of each face image is generated, and the target subject face image is identified from multiple face images, which can greatly improve the accuracy of subject face recognition.

[0224] Figure 3 FIG. 1 is a block diagram of a subject face image recognition device according to an exemplary embodiment. Figure 3 , the device comprises:

[0225] The face recognition module 310 is configured to perform face recognition on a target image including multiple faces to obtain multiple face images;

[0226] The multi-dimensional quality analysis module 320 is configured to perform multi-dimensional quality analysis on multiple face images to obtain initial quality data of each face image under multiple analysis dimensions;

[0227] The constraint processing module 330 is configured to execute the preset constraint quality data corresponding to each analysis dimension, perform constraint processing on the initial quality data under each analysis dimension, and obtain the constraint quality result of each face image under each analysis dimension. The preset constraint quality data is used to balance the initial quality data under each analysis dimension and the degree of influence on the quality of each face image.

[0228] The target quality data generating module 340 is configured to generate target quality data for each face image based on the constraint quality results of each face image under multiple analysis dimensions;

[0229] The target subject face image recognition module 350 is configured to identify a target subject face image from a plurality of face images based on the target quality data.

[0230] Optionally, the target subject face image recognition module 350 includes:

[0231] A preliminary subject face image recognition unit is configured to determine a preliminary subject face image corresponding to the maximum target quality data from a plurality of face images;

[0232] The target subject face image recognition unit is configured to execute the process of using the preliminary selected subject face image as the target subject face image.

[0233] Optionally, the target image is a preset frame image in a preset video, the primary subject face image is the target face image, and the above device further includes:

[0234] A first subject face image acquisition module is configured to execute acquisition of subject face images corresponding to a first number of frame images before a preset frame image;

[0235] The target subject face image recognition unit is further configured to perform:

[0236] In the case that the subject facial images corresponding to the first number of frame images include images of the second number of target faces, the preliminarily selected subject facial images are used as target subject facial images.

[0237] Optionally, the above device further includes:

[0238] A second subject face image acquisition module is configured to execute acquisition of subject face images corresponding to a third number of frame images before the preset frame image;

[0239] The target subject face image recognition unit is further configured to perform:

[0240] When the subject facial images corresponding to the first number of frame images include images of the second number of target faces, and the subject facial images corresponding to the third number of frame images are all images of the target faces, the preliminarily selected subject facial images are used as target subject facial images.

[0241] Optionally, the target image is a preset frame image in a preset video, and the target quality data generating module 340 includes:

[0242] an intermediate quality data determination unit, configured to determine the intermediate quality data of each face image according to the constraint quality results of each face image under multiple analysis dimensions;

[0243] A target face image determination unit is configured to determine a target face image including a historical subject face from a plurality of face images, wherein the historical subject face is a face in a subject face image corresponding to a historical frame image of the target image;

[0244] A scaling processing unit is configured to perform scaling processing on the intermediate quality data of the target face image based on a preset scale factor to obtain target quality data of the target face image;

[0245] The target quality data determination unit is configured to use the intermediate quality data of other human face images as the target quality data of other human face images, where the other human face images are human face images other than the target human face image among the multiple human face images.

[0246] Optionally, the target quality data of the target face image includes reduced quality data or enlarged quality data; and the scaling processing unit includes:

[0247] A preset scale factor determination unit is configured to determine a preset scale factor based on target scene information;

[0248] A reduction processing unit is configured to perform reduction processing on the intermediate quality data of the target face image based on the preset scale factor when the preset scale factor is a negative number, so as to obtain reduced quality data;

[0249] The enlargement processing unit is configured to perform enlargement processing on the intermediate quality data of the target face image based on the preset scale coefficient when the preset scale coefficient is a positive number, so as to obtain enlarged quality data.

[0250] Optionally, the preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes target constraint quality data;

[0251] The constraint processing module 330 includes:

[0252] The first target constraint quality data determination unit is configured to execute, in the case where the multiple face images include the first face image, using the upper limit of the quality data corresponding to the target analysis dimension as the target constraint quality data of the first face image under the target analysis dimension; the target analysis dimension is any analysis dimension among the multiple analysis dimensions, and the first face image is a face image among the multiple face images, the initial quality data under the target analysis dimension being greater than the upper limit of the quality data corresponding to the target analysis dimension;

[0253] The second target constraint quality data determination unit is configured to execute, when the plurality of face images include a second face image, using the lower limit of the quality data corresponding to the target analysis dimension as the target constraint quality data of the second face image under the target analysis dimension; the second face image is a face image among the plurality of face images, the initial quality data under the target analysis dimension of which is less than the lower limit of the quality data corresponding to the target analysis dimension;

[0254] The third target constraint quality data determination unit is configured to execute, when multiple facial images include a third facial image, using the initial quality data of the third facial image under the target analysis dimension as the target constraint quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the multiple facial images, whose initial quality data under the target analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension.

[0255] Optionally, the preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes target constraint quality data;

[0256] The constraint processing module 330 includes:

[0257] The first initial constrained quality data determination unit is configured to execute, in the case where the multiple facial images include the first facial image, using the upper limit of the quality data corresponding to the target analysis dimension as the initial constrained quality data of the first facial image under the target analysis dimension; the target analysis dimension is any analysis dimension among the multiple analysis dimensions, and the first facial image is a facial image among the multiple facial images, the initial quality data under the target analysis dimension being greater than the upper limit of the quality data corresponding to the target analysis dimension;

[0258] The second initial constrained quality data determining unit is configured to, when the plurality of face images include a second face image, use the lower limit of the quality data corresponding to the target analysis dimension as the initial constrained quality data of the second face image under the target analysis dimension; the second face image is a face image among the plurality of face images, the initial quality data under the target analysis dimension of which is less than the lower limit of the quality data corresponding to the target analysis dimension;

[0259] The third initial constrained quality data determining unit is configured to execute, in the case where the multiple facial images include a third facial image, using the initial quality data of the third facial image under the target analysis dimension as the initial constrained quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the multiple facial images, the initial quality data of which under the target analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension;

[0260] The target constraint quality data generating unit is configured to generate target constraint quality data for each face image in each analysis dimension based on the initial constraint quality data.

[0261] Optionally, the constraint quality result also includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality;

[0262] The constraint processing module 330 also includes:

[0263] A first quality determination result determining unit is configured to perform the quality determination result of the second face image as unqualified quality;

[0264] The second quality determination result determining unit is configured to use qualified quality as the quality determination result of the first face image and the three face images.

[0265] Optionally, the constraint quality result includes target constraint quality data;

[0266] The target quality data generation module 340 includes:

[0267] The weighted summation unit is configured to execute preset quality weights corresponding to multiple analysis dimensions, perform weighted summation on the target constraint quality data of each face image under multiple analysis dimensions, and obtain target quality data.

[0268] Optionally, the constraint quality result also includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality;

[0269] The target quality data generation module 340 also includes:

[0270] A key analysis dimension determination unit is configured to determine a key analysis dimension among a plurality of analysis dimensions;

[0271] An image filtering unit is configured to perform quality determination results based on key analysis dimensions, filter multiple face images, and obtain face images to be screened;

[0272] The weighted summation unit is also configured to perform:

[0273] Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of the face image to be screened under the multiple analysis dimensions to obtain the target quality data;

[0274] The target subject facial image recognition module 350 is further configured to identify the target subject facial image from the facial images to be screened based on the target quality data.

[0275] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0276] Figure 4 is a block diagram of an electronic device for subject face image recognition according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 4 As shown. The electronic device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a subject face image recognition method is implemented. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the housing of the electronic device, or an external keyboard, touchpad or mouse, etc.

[0277] Those skilled in the art will understand that Figure 4 The structure shown in the figure is merely a block diagram of a partial structure related to the scheme of the present disclosure, and does not constitute a limitation on the electronic device to which the scheme of the present disclosure is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0278] In an exemplary embodiment, an electronic device is also provided, including: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the subject face image recognition method as in the embodiment of the present disclosure.

[0279] In an exemplary embodiment, a computer-readable storage medium is also provided. When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the subject face image recognition method in the embodiment of the present disclosure.

[0280] In an exemplary embodiment, a computer program product including instructions is also provided. When the computer program product is run on a computer, the computer is enabled to execute the subject face image recognition method in the embodiment of the present disclosure.

[0281] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0282] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0283] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A subject face image recognition method, characterized in that: include: Performing face recognition on a target image including multiple faces to obtain multiple face images; Performing multi-dimensional quality analysis on the multiple facial images to obtain initial quality data of each facial image in multiple analysis dimensions; the multiple analysis dimensions include at least two of a facial size analysis dimension, a facial orientation analysis dimension, a facial position analysis dimension, and a gaze tracking analysis dimension; Based on the preset constraint quality data corresponding to each analysis dimension, the initial quality data under each analysis dimension is constrained to obtain the constrained quality result of each face image under each analysis dimension, wherein the preset constraint quality data is used to balance the initial quality data under each analysis dimension and the degree of influence on the quality of each face image; generating target quality data for each face image based on the constrained quality results of each face image under the multiple analysis dimensions; The constraint quality result includes target constraint quality data; The target constraint quality data is a parameter used to measure the quality of each face image under the multiple analysis dimensions after balancing the initial quality data under the multiple analysis dimensions and affecting the quality of each face image; the target quality data is used to measure the quality of each face image; Based on the target quality data, a target subject face image is identified from the multiple face images.

2. The subject face image recognition method according to claim 1, characterized in that: The step of identifying a target subject face image from the plurality of face images based on the target quality data comprises: Determining, from the plurality of face images, a primary subject face image corresponding to the maximum target quality data; The preliminarily selected subject face image is used as the target subject face image.

3. The subject face image recognition method according to claim 2, characterized in that: The target image is a preset frame image in a preset video, the primary subject face image is an image of the target face, and the method further includes: Acquire subject face images corresponding to a first number of frame images before the preset frame image; The using the pre-selected subject face image as the target subject face image comprises: In a case where the subject facial images corresponding to the first number of frame images include images of the second number of target faces, the preliminarily selected subject facial images are used as the target subject facial images.

4. The subject face image recognition method according to claim 3, characterized in that: The method further comprises: Acquire subject face images corresponding to a third number of frame images before the preset frame image; In the case that the subject face images corresponding to the first number of frame images include the second number of target face images, using the preliminarily selected subject face images as the target subject face images comprises: When the subject facial images corresponding to the first number of frame images include images of the second number of the target faces, and the subject facial images corresponding to the third number of frame images are all images of the target faces, the preliminarily selected subject facial images are used as the target subject facial images.

5. The subject face image recognition method according to claim 1, characterized in that: The target image is a preset frame image in a preset video, and generating target quality data of each face image based on the constraint quality results of each face image under the multiple analysis dimensions includes: Determining intermediate quality data of each face image according to the constraint quality results of each face image under the multiple analysis dimensions; Determine a target face image including a historical subject face from the plurality of face images, wherein the historical subject face is a face in a subject face image corresponding to a historical frame image of the target image; Based on a preset scaling factor, scaling the intermediate quality data of the target face image to obtain target quality data of the target face image; The intermediate quality data of other human face images are used as the target quality data of the other human face images, and the other human face images are human face images other than the target human face image among the multiple human face images.

6. The subject face image recognition method according to claim 5, characterized in that: The target quality data of the target facial image includes reduced quality data or enlarged quality data; the target quality data of the target facial image obtained by scaling the intermediate quality data of the target facial image based on a preset scale factor includes: Determining the preset scaling factor based on the target scene information; When the preset scale factor is a negative number, the intermediate quality data of the target face image is reduced based on the preset scale factor to obtain the reduced quality data; When the preset proportionality coefficient is a positive number, the intermediate quality data of the target face image is enlarged based on the preset proportionality coefficient to obtain the enlarged quality data.

7. The subject face image recognition method according to claim 1, characterized in that: The preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes a target constraint quality data; The constraining process is performed on the initial quality data under each analysis dimension based on the preset constrained quality data corresponding to each analysis dimension to obtain the constrained quality result of each face image under each analysis dimension, including: In a case where the multiple facial images include a first facial image, using an upper limit of quality data corresponding to a target analysis dimension as target constraint quality data of the first facial image under the target analysis dimension; the target analysis dimension is any analysis dimension among the multiple analysis dimensions, and the first facial image is a facial image among the multiple facial images, the initial quality data of which under the target analysis dimension is greater than the upper limit of quality data corresponding to the target analysis dimension; In the case where the multiple facial images include a second facial image, the lower limit of the quality data corresponding to the target analysis dimension is used as the target constraint quality data of the second facial image under the target analysis dimension; the second facial image is a facial image among the multiple facial images, the initial quality data under the target analysis dimension of which is less than the lower limit of the quality data corresponding to the target analysis dimension; In the case that the multiple facial images include a third facial image, the initial quality data of the third facial image under the target analysis dimension is used as the target constraint quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the multiple facial images, whose initial quality data under the target analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension.

8. The subject face image recognition method according to claim 1, characterized in that: The preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes a target constraint quality data; The constraining process is performed on the initial quality data under each analysis dimension based on the preset constrained quality data corresponding to each analysis dimension to obtain the constrained quality result of each face image under each analysis dimension, including: In a case where the multiple facial images include a first facial image, using an upper limit of quality data corresponding to a target analysis dimension as initial constrained quality data of the first facial image under the target analysis dimension; the target analysis dimension is any analysis dimension among the multiple analysis dimensions, and the first facial image is a facial image among the multiple facial images, the initial quality data of which under the target analysis dimension is greater than an upper limit of quality data corresponding to the target analysis dimension; In the case where the multiple facial images include a second facial image, the lower limit of the quality data corresponding to the target analysis dimension is used as the initial constraint quality data of the second facial image under the target analysis dimension; the second facial image is a facial image among the multiple facial images whose initial quality data under the target analysis dimension is less than the lower limit of the quality data corresponding to the target analysis dimension; In the case where the multiple facial images include a third facial image, initial quality data of the third facial image under the target analysis dimension is used as initial constrained quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the multiple facial images whose initial quality data under the target analysis dimension is greater than or equal to a lower limit of quality data corresponding to the target analysis dimension and less than an upper limit of quality data corresponding to the target analysis dimension; Based on the initial constraint quality data, target constraint quality data of each face image in each analysis dimension is generated.

9. The subject face image recognition method according to claim 7 or 8, characterized in that: The constraint quality result also includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality; The constraining the initial quality data under each analysis dimension based on the preset constrained quality data corresponding to each analysis dimension to obtain the constrained quality result of each face image under each analysis dimension also includes: Taking unqualified quality as the quality determination result of the second facial image; The qualified quality is used as the quality judgment result of the first face image and the three face images.

10. The subject face image recognition method according to any one of claims 1 to 8, characterized in that: The constraint quality result includes target constraint quality data; The generating target quality data of each face image based on the constraint quality results of each face image under the multiple analysis dimensions comprises: Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of each face image under the multiple analysis dimensions to obtain the target quality data.

11. The subject face image recognition method according to claim 10, characterized in that: The constraint quality result also includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality; The generating target quality data of each face image based on the constraint quality results of each face image under the multiple analysis dimensions further comprises: determining a key analysis dimension among the plurality of analysis dimensions; Based on the quality determination results under the key analysis dimensions, the multiple face images are filtered to obtain face images to be screened; The weighted summation of the target constraint quality data of each face image under the multiple analysis dimensions based on the preset quality weights corresponding to the multiple analysis dimensions to obtain the target quality data includes: Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of the face image to be screened under the multiple analysis dimensions to obtain the target quality data; The step of identifying a target subject face image from the plurality of face images based on the target quality data comprises: Based on the target quality data, the target subject face image is identified from the face images to be screened.

12. A subject face image recognition device, characterized in that: include: A face recognition module is configured to perform face recognition on a target image including multiple faces to obtain multiple face images; a multi-dimensional quality analysis module configured to perform multi-dimensional quality analysis on the plurality of face images to obtain initial quality data of each face image under a plurality of analysis dimensions; the plurality of analysis dimensions comprising at least two of a face size analysis dimension, a face orientation analysis dimension, a face position analysis dimension, and a gaze tracking analysis dimension; a constraint processing module configured to execute, based on preset constraint quality data corresponding to each analysis dimension, constraint processing on the initial quality data under each analysis dimension to obtain the constraint quality result of each face image under each analysis dimension, wherein the preset constraint quality data is used to balance the initial quality data under each analysis dimension and the degree of influence on the quality of each face image; A target quality data generating module is configured to generate target quality data for each face image based on the constraint quality results of each face image under the multiple analysis dimensions; The constraint quality result includes target constraint quality data; The target constraint quality data is a parameter used to measure the quality of each face image under the multiple analysis dimensions after balancing the initial quality data under the multiple analysis dimensions and affecting the quality of each face image; the target quality data is used to measure the quality of each face image; The target subject facial image recognition module is configured to identify a target subject facial image from the multiple facial images based on the target quality data.

13. The subject face image recognition device according to claim 12, characterized in that: The target subject face image recognition module includes: A preliminary subject face image recognition unit is configured to determine a preliminary subject face image corresponding to the maximum target quality data from the plurality of face images; The target subject face image recognition unit is configured to execute the process of using the preliminary selected subject face image as the target subject face image.

14. The subject face image recognition device according to claim 13, characterized in that: The target image is a preset frame image in a preset video, the pre-selected subject face image is an image of the target face, and the device further includes: A first subject face image acquisition module is configured to execute acquisition of subject face images corresponding to a first number of frame images before the preset frame image; The target subject face image recognition unit is further configured to perform: In a case where the subject facial images corresponding to the first number of frame images include images of the second number of target faces, the preliminarily selected subject facial images are used as the target subject facial images.

15. The subject face image recognition device according to claim 14, characterized in that: The device also includes: A second subject face image acquisition module is configured to execute acquisition of subject face images corresponding to a third number of frame images before the preset frame image; The target subject face image recognition unit is further configured to perform: When the subject facial images corresponding to the first number of frame images include images of the second number of the target faces, and the subject facial images corresponding to the third number of frame images are all images of the target faces, the preliminarily selected subject facial images are used as the target subject facial images.

16. The subject face image recognition device according to claim 12, characterized in that: The target image is a preset frame image in a preset video, and the target quality data generation module includes: an intermediate quality data determining unit, configured to determine the intermediate quality data of each face image according to the constraint quality results of each face image under the multiple analysis dimensions; A target facial image determining unit is configured to determine a target facial image including a historical subject face from the plurality of facial images, wherein the historical subject face is a face in a subject facial image corresponding to a historical frame image of the target image; A scaling processing unit is configured to perform scaling processing on the intermediate quality data of the target face image based on a preset scaling coefficient to obtain target quality data of the target face image; The target quality data determination unit is configured to use the intermediate quality data of other human face images as the target quality data of the other human face images, wherein the other human face images are human face images other than the target human face image among the multiple human face images.

17. The subject face image recognition device according to claim 16, characterized in that: The target quality data of the target face image includes reduced quality data or enlarged quality data; the scaling processing unit includes: A preset scale factor determination unit is configured to determine the preset scale factor based on target scene information; a reduction processing unit configured to perform reduction processing on the intermediate quality data of the target face image based on the preset scale factor when the preset scale factor is a negative number, so as to obtain the reduced quality data; The enlargement processing unit is configured to perform enlargement processing on the intermediate quality data of the target face image based on the preset scale factor when the preset scale factor is a positive number, so as to obtain the enlarged quality data.

18. The subject face image recognition device according to claim 12, characterized in that: The preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes a target constraint quality data; The constraint processing module comprises: The first target constraint quality data determination unit is configured to, when the plurality of face images include a first face image, use an upper limit of quality data corresponding to a target analysis dimension as target constraint quality data of the first face image under the target analysis dimension; the target analysis dimension is any analysis dimension among the plurality of analysis dimensions, and the first face image is a face image among the plurality of face images, the initial quality data under the target analysis dimension being greater than the upper limit of quality data corresponding to the target analysis dimension; The second target constraint quality data determination unit is configured to use, when the plurality of face images include a second face image, a lower limit of quality data corresponding to the target analysis dimension as the target constraint quality data of the second face image under the target analysis dimension; the second face image is a face image among the plurality of face images, the initial quality data under the target analysis dimension of which is less than the lower limit of quality data corresponding to the target analysis dimension; The third target constraint quality data determination unit is configured to execute, when the multiple facial images include a third facial image, using the initial quality data of the third facial image under the target analysis dimension as the target constraint quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the multiple facial images, whose initial quality data under the target analysis dimension is greater than or equal to the lower limit of the quality data corresponding to the target analysis dimension, and less than the upper limit of the quality data corresponding to the target analysis dimension.

19. The subject face image recognition device according to claim 12, characterized in that: The preset constraint quality data includes a quality data upper limit and a quality data lower limit; the constraint quality result includes a target constraint quality data; The constraint processing module comprises: The first initial constrained quality data determination unit is configured to, when the plurality of facial images include a first facial image, use an upper limit of quality data corresponding to a target analysis dimension as initial constrained quality data of the first facial image under the target analysis dimension; the target analysis dimension is any analysis dimension among the plurality of analysis dimensions, and the first facial image is a facial image among the plurality of facial images, the initial quality data under the target analysis dimension being greater than an upper limit of quality data corresponding to the target analysis dimension; The second initial constrained quality data determining unit is configured to, when the plurality of facial images include a second facial image, use the lower limit of the quality data corresponding to the target analysis dimension as the initial constrained quality data of the second facial image under the target analysis dimension; the second facial image is a facial image among the plurality of facial images whose initial quality data under the target analysis dimension is less than the lower limit of the quality data corresponding to the target analysis dimension; A third initial constrained quality data determination unit is configured to, when the plurality of facial images include a third facial image, use the initial quality data of the third facial image under the target analysis dimension as the initial constrained quality data of the third facial image under the target analysis dimension; the third facial image is a facial image among the plurality of facial images, the initial quality data of which under the target analysis dimension is greater than or equal to a lower limit of the quality data corresponding to the target analysis dimension and less than an upper limit of the quality data corresponding to the target analysis dimension; The target constraint quality data generating unit is configured to generate the target constraint quality data of each face image in each analysis dimension based on the initial constraint quality data.

20. The subject face image recognition device according to claim 18 or 19, characterized in that: The constraint quality result also includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality; The constraint processing module also includes: A first quality determination result determining unit is configured to perform unqualified quality determination as a quality determination result of the second facial image; The second quality determination result determining unit is configured to use qualified quality as the quality determination result of the first face image and the three-person face images.

21. The subject face image recognition device according to any one of claims 12 to 19, characterized in that: The constraint quality result includes target constraint quality data; The target quality data generation module includes: The weighted summation unit is configured to perform weighted summation of the target constraint quality data of each face image under the multiple analysis dimensions based on preset quality weights corresponding to the multiple analysis dimensions to obtain the target quality data.

22. The subject face image recognition device according to claim 21, characterized in that: The constraint quality result also includes a quality determination result, and the quality determination result includes qualified quality and / or unqualified quality; The target quality data generation module also includes: A key analysis dimension determination unit, configured to determine a key analysis dimension among the plurality of analysis dimensions; An image filtering unit is configured to perform filtering on the plurality of face images based on the quality determination result under the key analysis dimension to obtain a face image to be screened; The weighted summation unit is further configured to perform: Based on the preset quality weights corresponding to the multiple analysis dimensions, weighted summation is performed on the target constraint quality data of the face image to be screened under the multiple analysis dimensions to obtain the target quality data; The target subject facial image recognition module is further configured to identify the target subject facial image from the facial images to be screened based on the target quality data.

23. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the subject face image recognition method as described in any one of claims 1 to 11.

24. A computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the subject face image recognition method as described in any one of claims 1 to 11.

25. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the subject face image recognition method described in any one of claims 1 to 11 is implemented.

Images