A face image quality evaluation method, device and electronic equipment

Abstract

The present disclosure provides a face image quality evaluation method and device and electronic equipment. The face quality feature of a face image to be evaluated is extracted. High-quality face features and poor-quality face features are generated according to a sampled normal distribution vector. The variance entropy between the face quality feature and the high-quality face feature is determined. The relative entropy between the face quality feature and the high-quality face feature, the poor-quality face feature, and the distribution entropy between the high-quality face feature and the poor-quality face feature are determined. The training feature value is obtained by fusing the above entropy values. The target variance between the face quality feature and the high-quality face feature is determined by a target face quality evaluator with minimized training feature value, and compared with a variance threshold to determine the image quality of the face image to be evaluated. The face image quality can be accurately judged and distinguished comprehensively.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of face recognition, in particular, to a face image quality evaluation method and device and electronic equipment. BACKGROUND

[0002] Face recognition is a biometric technology based on face feature information for identity recognition. It uses a camera or camera to collect images or video streams containing faces, and automatically detects and tracks faces in the images, and then performs a series of related technologies for face recognition. The quality of the face image determines the accuracy of face detection, face liveness distinction and face recognition. The recognition accuracy of a high-quality face image is much higher than that of a poor-quality face image. Therefore, how to evaluate the quality of a face image is very important in the application of a face recognition system.

[0003] In the existing face image quality evaluation process, the commonly used methods usually include brightness judgment, sharpness judgment, blur judgment and the like. However, the above-mentioned face image quality evaluation methods have limitations and cannot accurately and comprehensively distinguish the quality of a face image, and the evaluation accuracy is low. SUMMARY

[0004] The embodiments of the present disclosure provide at least a face image quality evaluation method, device and electronic equipment, which can accurately distinguish the quality of a face image.

[0005] The embodiments of the present disclosure provide a face image quality evaluation method, which comprises:

[0006] Obtaining a face image to be evaluated, and extracting face quality features corresponding to the face image to be evaluated;

[0007] Sampling a set of normal distribution vectors conforming to a normal distribution in a face feature space, and generating high-quality face features and poor-quality face features in the face feature space according to the set of normal distribution vectors;

[0008] Determining a first correlation degree between the face quality features and the high-quality face features, and determining a variance entropy between the face quality features and the high-quality face features according to the first correlation degree by a preset face quality evaluator;

[0009] Determining a second correlation degree between the face quality features and the poor-quality face features, and determining a relative entropy between the face quality features and the high-quality face features and the poor-quality face features according to the first correlation degree and the second correlation degree;

[0010] determine a third correlation degree between the high-quality face feature and the low-quality face feature, and determine a distribution entropy between the high-quality face feature and the low-quality face feature according to the third correlation degree;

[0011] fuse the variance entropy, the relative entropy and the distribution entropy to generate a training feature value corresponding to the face quality evaluator, and determine a target face quality evaluator that minimizes the training feature value;

[0012] determine a target variance between the face quality feature and the high-quality face feature by using the target face quality evaluator, and determine the image quality of the face image to be evaluated based on the target variance and a preset variance threshold.

[0013] In an optional implementation, the normal distribution vector set that conforms to the normal distribution in the face feature space is sampled, and the high-quality face feature and the low-quality face feature in the face feature space are generated according to the normal distribution vector set, and specifically includes:

[0014] each element in the normal distribution vector set is input into a preset high-quality face feature generator, and a high-quality face feature set in the face feature space is determined, wherein the high-quality face feature set is composed of a plurality of high-quality face features;

[0015] each element in the normal distribution vector set is input into a preset low-quality face feature generator, and a low-quality face feature set in the face feature space is determined, wherein the low-quality face feature set is composed of a plurality of low-quality face features.

[0016] In an optional implementation, the first correlation degree between the face quality feature and the high-quality face feature is determined, and a variance entropy between the face quality feature and the high-quality face feature is determined according to the first correlation degree by using a preset face quality evaluator, and specifically includes:

[0017] the variance between the face quality feature and the high-quality face feature is determined according to the first correlation degree by using the face quality evaluator;

[0018] the variance entropy between the face quality feature and the high-quality face feature is determined according to the variance.

[0019] In an optional implementation, after the second correlation degree between the face quality feature and the low-quality face feature is determined, and the relative entropy between the face quality feature and the high-quality face feature and the low-quality face feature is determined according to the first correlation degree and the second correlation degree, the method further includes:

[0020] perform partial differential operation on the poor-quality face feature based on the relative entropy, to optimize accuracy of the poor-quality face feature generated by the poor-quality feature generator.

[0021] perform partial differential operation on the poor-quality face feature based on the relative entropy, to optimize accuracy of the poor-quality face feature generated by the poor-quality feature generator.

[0022] In an optional implementation, the variance entropy, the relative entropy, and the distribution entropy are fused based on the following method:

[0023] corresponding preset weight coefficients are configured for the variance entropy, the relative entropy, and the distribution entropy, respectively;

[0024] the variance entropy, the relative entropy, and the distribution entropy are weighted and summed according to the preset weight coefficients, to obtain a training feature value corresponding to the face quality evaluator.

[0025] In an optional implementation, the image quality of the face image to be evaluated is determined based on the target variance and a preset variance threshold, and specifically includes:

[0026] determining whether the target variance is less than the variance threshold;

[0027] if less, determining that the face image to be evaluated is a high-quality image;

[0028] if greater, determining that the face image to be evaluated is a poor-quality image.

[0029] The embodiments of the present disclosure further provide a device for evaluating image quality of a face image, and the device includes:

[0030] a feature extraction module configured to acquire a face image to be evaluated, and extract a face quality feature corresponding to the face image to be evaluated;

[0031] a feature generation module configured to sample a set of normal distribution vectors conforming to a normal distribution in a face feature space, and generate high-quality face features and poor-quality face features in the face feature space according to the set of normal distribution vectors;

[0032] a variance entropy determination module configured to determine a first correlation degree between the face quality feature and the high-quality face feature, and determine a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree by using a preset face quality evaluator;

[0033] The relative entropy determination module is configured to determine a second correlation degree between the face quality feature and the poor-quality face feature, and determine a relative entropy between the face quality feature, the high-quality face feature and the poor-quality face feature according to the first correlation degree and the second correlation degree.

[0034] The distribution entropy determination module is configured to determine a third correlation degree between the high-quality face feature and the poor-quality face feature, and determine a distribution entropy between the high-quality face feature and the poor-quality face feature according to the third correlation degree.

[0035] The training module is configured to generate a training feature value corresponding to the face quality evaluator by fusing the variance entropy, the relative entropy and the distribution entropy, and determine a target face quality evaluator that minimizes the training feature value.

[0036] The image quality evaluation module is configured to determine a target variance between the face quality feature and the high-quality face feature by using the target face quality evaluator, and determine an image quality of the face image to be evaluated based on the target variance and a preset variance threshold.

[0037] In an optional implementation, the feature generation module is specifically configured to:

[0038] input each element in the set of normal distribution vectors into a preset high-quality face feature generator, and determine a set of high-quality face features in the face feature space, wherein the set of high-quality face features is composed of a plurality of high-quality face features;

[0039] input each element in the set of normal distribution vectors into a preset poor-quality face feature generator, and determine a set of poor-quality face features in the face feature space, wherein the set of poor-quality face features is composed of a plurality of poor-quality face features.

[0040] In an optional implementation, the variance entropy determination module is specifically configured to:

[0041] determine, by the face quality evaluator, a variance between the face quality feature and the high-quality face feature according to the first correlation degree;

[0042] determine a variance entropy between the face quality feature and the high-quality face feature according to the variance.

[0043] In an optional implementation, the training module is specifically configured to:

[0044] configure a preset weight coefficient corresponding to each of the variance entropy, the relative entropy and the distribution entropy;

[0045] According to the preset weight coefficient, the variance entropy, the relative entropy and the distribution entropy are weighted and summed to obtain a training feature value corresponding to the face quality evaluator.

[0046] In an alternative embodiment, the image quality evaluation module is specifically configured to:

[0047] determine whether the target variance is less than the variance threshold value;

[0048] if less than, determine that the face image to be evaluated is a high-quality image;

[0049] if greater than, determine that the face image to be evaluated is a low-quality image.

[0050] In an alternative embodiment, the device further comprises an optimization module configured to:

[0051] based on the relative entropy, perform partial differential operation on the high-quality face feature to optimize the accuracy of the high-quality feature generator in generating the high-quality face feature;

[0052] based on the relative entropy, perform partial differential operation on the low-quality face feature to optimize the accuracy of the low-quality feature generator in generating the low-quality face feature.

[0053] The embodiments of the present disclosure further provide an electronic device, which comprises a processor, a memory and a bus, the memory stores machine readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, and the machine readable instructions are executed by the processor to perform the face image quality evaluation method or the steps in any one of the possible embodiments of the face image quality evaluation method.

[0054] The embodiments of the present disclosure further provide a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to perform the face image quality evaluation method or the steps in any one of the possible embodiments of the face image quality evaluation method.

[0055] The embodiments of the present disclosure further provide a computer program product, which comprises a computer program / instruction, and the computer program / instruction is executed by a processor to implement the face image quality evaluation method or the steps in any one of the possible embodiments of the face image quality evaluation method.

[0056] The method, device and electronic equipment for evaluating face image quality provided by the embodiments of the present disclosure can comprehensively and accurately judge and distinguish the face image quality by obtaining a face image to be evaluated, extracting a face quality feature corresponding to the face image to be evaluated, sampling a normal distribution vector set conforming to a normal distribution in a face feature space, generating a high-quality face feature and a poor-quality face feature in the face feature space according to the normal distribution vector set, determining a first correlation degree between the face quality feature and the high-quality face feature, determining a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree by a preset face quality evaluator, determining a second correlation degree between the face quality feature and the poor-quality face feature, determining a relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature according to the first correlation degree and the second correlation degree, determining a third correlation degree between the high-quality face feature and the poor-quality face feature, determining a distribution entropy between the high-quality face feature and the poor-quality face feature according to the third correlation degree, generating a training characteristic value corresponding to the face quality evaluator by fusing the variance entropy, the relative entropy and the distribution entropy, determining a target face quality evaluator that minimizes the training characteristic value, and determining an image quality of the face image to be evaluated based on a target variance between the face quality feature and the high-quality face feature determined by the target face quality evaluator and a preset variance threshold.

[0057] In order to make the above objectives, features and advantages of the present disclosure more obvious and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are described in detail as follows. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following will briefly introduce the drawings needed to be used in the embodiments. The drawings herein are incorporated into the specification and form a part of the specification, which illustrate the embodiments consistent with the present disclosure, and are used to explain the technical solutions of the present disclosure together with the specification. It should be understood that the following drawings only show some embodiments of the present disclosure, and therefore should not be considered as a limitation to the scope, and other related drawings can also be obtained by those skilled in the art without creative labor.

[0059] Figure 1 A flowchart of a method for evaluating face image quality provided by the embodiments of the present disclosure is shown;

[0060] Figure 2 A flowchart of a feature generator optimization method provided by the embodiments of the present disclosure is shown;

[0061] Figure 3 A schematic diagram of an evaluation device for face image quality provided by the embodiments of the present disclosure is shown;

[0062] Figure 4 A schematic diagram of an electronic device is shown. DETAILED DESCRIPTION

[0063] To make the objectives, technical solutions and advantages of the embodiments of the present disclosure clearer, the following will be combined with the accompanying drawings for the embodiments of the present disclosure to make a clear and complete description of the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. The components of the embodiments of the present disclosure generally described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed present disclosure, but only represents selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0064] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0065] The term "and / or" herein is only to describe an association relationship, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. In addition, the term "at least one" herein means any one of a plurality or any combination of at least two of a plurality, for example, including at least one of A, B and C can mean including any one or more elements selected from the set consisting of A, B and C.

[0066] It is found through research that in the existing face image quality evaluation process, the commonly used methods usually include brightness judgment, sharpness judgment, blur judgment and the like. However, the above face image quality evaluation methods have limitations and cannot accurately and comprehensively judge and distinguish the face image quality, and the evaluation accuracy is low.

[0067] Based on the above research, the present disclosure provides a face image quality evaluation method, device and electronic equipment. The face image quality evaluation method comprises the following steps: obtaining a face image to be evaluated, and extracting a face quality feature corresponding to the face image to be evaluated; sampling a normal distribution vector set conforming to a normal distribution in a face feature space, and generating a high-quality face feature and a poor-quality face feature in the face feature space according to the normal distribution vector set; determining a first correlation degree between the face quality feature and the high-quality face feature, determining a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree by using a preset face quality evaluator; determining a second correlation degree between the face quality feature and the poor-quality face feature, and determining a relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature according to the first correlation degree and the second correlation degree; determining a third correlation degree between the high-quality face feature and the poor-quality face feature, and determining a distribution entropy between the high-quality face feature and the poor-quality face feature according to the third correlation degree; fusing the variance entropy, the relative entropy and the distribution entropy to generate a training characteristic value corresponding to the face quality evaluator, and determining a target face quality evaluator that minimizes the training characteristic value; determining a target variance between the face quality feature and the high-quality face feature by using the target face quality evaluator, and determining an image quality of the face image to be evaluated based on the target variance and a preset variance threshold. The face image quality can be accurately judged and distinguished comprehensively.

[0068] To facilitate the understanding of the present embodiment, first, a face image quality evaluation method disclosed by the present embodiment is introduced in detail. The execution subject of the face image quality evaluation method provided by the present embodiment is generally a computer device with certain computing capability, which may, for example, include a terminal device or a server or other processing device. The terminal device may be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementation manners, the face image quality evaluation method may be realized by a processor calling computer readable instructions stored in a memory.

[0069] Referring to Figure 1 FIG. 1 shows a flowchart of a face image quality evaluation method provided by the present embodiment. The method comprises steps S101-S107, wherein:

[0070] S101, obtaining a face image to be evaluated, and extracting a face quality feature corresponding to the face image to be evaluated.

[0071] In a specific implementation, a face image for face recognition is first acquired, which can be an image containing a face captured by a camera or a camera head or the like, or a frame of video image in a video stream containing a face collected. In order to ensure the accuracy of the face recognition result, the image quality of the face image needs to be evaluated before being input into the face recognition system. Face recognition is performed on high-quality face images, which often has higher recognition accuracy than low-quality face images.

[0072] Therefore, the image containing a face captured by a camera or a camera head or the like, or a frame of video image in a video stream containing a face collected is taken as a face image to be evaluated, and a face quality feature reflecting the image quality of the face contained in the face image to be evaluated is extracted from the face image to be evaluated.

[0073] Here, the face quality feature is a feature reflecting the image quality of the face displayed in the face image to be evaluated, which can specifically include brightness features, exposure features, definition features, and blur features, and further, the face quality feature has a form of a string of numerical values.

[0074] The face quality feature can be extracted by a preset face quality feature extractor trained to extract quality features for faces. For the face image to be evaluated containing a face, the face quality feature can be obtained after the face image to be evaluated is input into the face quality feature extractor.

[0075] Specifically, after the face image to be evaluated x is input into the face quality feature extractor F(x), the face quality feature f=F(x) is obtained, wherein, w represents the width of the face image to be evaluated, and h represents the height of the face image to be evaluated; n represents the feature dimension of the face quality feature.

[0076] It should be noted that the face quality feature extractor can be constructed by any convolutional neural network, which can only complete the task of extracting the face quality feature after being trained. The type of the extracted face quality feature can be selected according to actual needs, which is not specifically limited here.

[0077] S102, a set of normal distribution vectors conforming to a normal distribution is sampled in a face feature space, and high-quality face features and low-quality face features in the face feature space are generated according to the set of normal distribution vectors.

[0078] In a specific implementation, multiple random normal distribution samplings are continuously performed in the face feature space to obtain a normal distribution vector set, for example, k times of normal distribution samplings are continuously performed in the face feature space to obtain a normal distribution vector set Y = {y1, y2, … yk}. k Then, the high-quality face features and the poor-quality face features in the face feature space are generated according to the normal distribution vector set obtained by the normal distribution sampling.

[0079] Here, the high-quality face features can be quality features corresponding to faces displayed in face images that are normal in brightness, normal in exposure, normal in definition, and free of blur, etc.; and the poor-quality face features can be quality features corresponding to faces displayed in face images that at least do not meet one of the evaluation indexes of brightness, exposure, definition, and blur, etc.

[0080] It should be noted that the distinguishing criteria of the high-quality face features and the poor-quality face features can be comprehensively determined according to the operation capability and network size of a subsequent face system, or manually set according to different task requirements, and in actual application, the selection is made according to actual needs, which is not specifically limited here.

[0081] Here, the normal distribution vector set includes multiple vectors randomly sampled in the face feature space and subject to normal distribution, and the number thereof is determined according to the number of normal distribution samplings, which can be selected as needed in actual application, which is not specifically limited here.

[0082] As a possible implementation, the high-quality face features and the poor-quality face features in the face feature space can be generated in the following manner: each element in the normal distribution vector set is input into a preset face high-quality feature generator to determine a high-quality face feature set in the face feature space, wherein the high-quality face feature set is composed of multiple high-quality face features; and each element in the normal distribution vector set is input into a preset face poor-quality feature generator to determine a poor-quality face feature set in the face feature space, wherein the poor-quality face feature set is composed of multiple poor-quality face features.

[0083] Here, the high-quality face features can be generated by using the preset face high-quality feature generator, which is trained to generate high-quality face features, for example, features that are normal in brightness, normal in exposure, normal in definition, and free of blur, etc. For each vector subject to normal distribution and randomly sampled in the face feature space in the normal distribution vector set, the high-quality face feature can be generated after the vector is input into the face high-quality feature generator.

[0084] Specifically, the vector y subject to normal distribution is input into the face high-quality feature generator G g(x), to generate a high-quality face feature y g = G g (y), wherein, m is the dimension of the vector y, and y ~ N(0, 1); y g ∈ R n n is the dimension of the high-quality face feature y g .

[0085] It should be noted that the face quality feature generator can be constructed by any convolutional neural network, as long as it can complete the task of generating high-quality face features after training.

[0086] Further, the poor-quality face feature can be generated by a preset face poor-quality feature generator. The face poor-quality feature generator is trained to generate poor-quality face features, such as features with abnormal brightness (too bright or too dark), abnormal exposure (overexposure or underexposure), low clarity, or blurring. For each vector randomly sampled in the face feature space and subject to a normal distribution in the normal distribution vector set, the poor-quality face feature can be generated after inputting the vector into the face poor-quality feature generator.

[0087] Specifically, after the normal distribution vector y is input into the face poor-quality feature generator G b (x), to generate a high-quality face feature y b = G b (y), wherein, m is the dimension of the vector y, and y ~ N(0, 1); y b ∈ R n n is the dimension of the high-quality face feature y b .

[0088] It should be noted that the face poor-quality feature generator can be constructed by any convolutional neural network, as long as it can complete the task of generating poor-quality face features after training.

[0089] S103, determining a first correlation degree between the face quality feature and the high-quality face feature, and determining, by a preset face quality evaluator, a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree.

[0090] In specific implementation, the face quality evaluator can determine the variance between the face quality feature and the high-quality face feature according to the first correlation degree between the face quality feature and the high-quality face feature, and then determine the variance entropy between the face quality feature and the high-quality face feature according to the variance.

[0091] Specifically, the first correlation degree between the face quality feature and the high-quality face feature can be calculated based on the following formula:

[0092]

[0093] wherein, σ(f, y g ) represents the first correlation degree between the face quality feature and the high-quality face feature; f represents the face quality feature; y g represents the high-quality face feature.

[0094] Further, the variance between the face quality feature and the high-quality face feature can be calculated based on the following formula:

[0095]

[0096] wherein, var represents the variance between the face quality feature and the high-quality face feature; E1 represents the face quality evaluator; i represents the face image to be evaluated; Y g represents the high-quality face feature; k represents the sampling number of normal distribution; σ(f(i), y jg ) represents the first correlation degree between the face quality feature F(i) and the high-quality face feature F represents the face quality feature extractor for extracting the face quality feature.

[0097] Further, the variance entropy between the face quality feature and the high-quality face feature can be represented as (2t-1)var, and it needs to be noted that, here, t=1 when the face image to be evaluated i is a high-quality image; t=0 when the face image to be evaluated i is a poor-quality image.

[0098] Here, the smaller the variance entropy between the face quality feature and the high-quality face feature, the more accurate the extraction of the face quality feature and the more accurate the generation of the high-quality face feature, and vice versa, the larger the variance entropy between the face quality feature and the high-quality face feature, the lower the accuracy of the extraction of the face quality feature and the lower the accuracy of the generation of the high-quality face feature.

[0099] S104, determining the second correlation degree between the face quality feature and the poor-quality face feature, and determining the relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature according to the first correlation degree and the second correlation degree.

[0100] In specific implementation, a preset face quality relative entropy evaluator can be used to determine the relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature according to the first correlation degree between the face quality feature and the high-quality face feature and the second correlation degree between the face quality feature and the poor-quality face feature.

[0101] It needs to be noted that, the preset face quality relative entropy evaluator can be built by any convolutional neural network.

[0102] Specifically, the relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature can be calculated based on the following formula:

[0103]

[0104] wherein, φ(i, y g , y b ) represents the relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature; i represents the face image to be evaluated; y g represents the high-quality face feature; y b represents the poor-quality face feature; k represents the number of normal distribution sampling; σ(F(i, y jg )) represents the first correlation degree; σ(F(i, y jb )) represents the second correlation degree; F represents the face quality feature extractor for extracting the face quality feature.

[0105] It should be noted that it is necessary to preset that when the face image to be evaluated i is a high-quality image, t = 1; when the face image to be evaluated i is a poor-quality image, t = 0. The calculation method of the second correlation degree can refer to the formula for calculating the first correlation degree in step S103, and will not be described here.

[0106] Here, the smaller the relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature, the more accurate the extraction of the face quality feature, the more accurate the generated high-quality face feature, and the more accurate the generated poor-quality face feature. On the contrary, the larger the relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature, the lower the accuracy of the extraction of the face quality feature, the lower the accuracy of the generated high-quality face feature, and the lower the accuracy of the generated poor-quality face feature.

[0107] Further, as described above in the manner of implementing step S102 by using the high-quality face feature generator and the poor-quality face feature generator, after step S104, the following steps S201-S202 can also be performed, as shown in the flowchart of the feature generator optimization method provided by the embodiment of the disclosure, the method comprising steps S201-S202, wherein: Figure 2

[0108] S201, based on the relative entropy, partial differential operation is performed on the high-quality face feature to optimize the accuracy of the high-quality feature generator in generating the high-quality face feature.

[0109] S202, based on the relative entropy, partial differential operation is performed on the poor-quality face feature to optimize the accuracy of the poor-quality feature generator in generating the poor-quality face feature.

[0110] ​In a specific implementation, by the above steps S201-S202, the relative entropy among the face quality feature, the high-quality face feature and the low-quality face feature is subjected to partial differential operation with respect to the high-quality face feature and the low-quality face feature, so as to gradually optimize and approximate the high-quality face feature and the low-quality face feature generated by the high-quality face feature generator and the low-quality face feature generator, and further optimize the high-quality face feature generator and the low-quality face feature generator.

[0111] Specifically, the relative entropy among the face quality feature, the high-quality face feature and the low-quality face feature is subjected to partial differential operation with respect to the high-quality face feature and the low-quality face feature based on the following formula:

[0112] wherein, represents a random initial value of iteration; represents the result of the first iteration; the subscript g|b of y represents g or b, and when the subscript of y is g, it represents the high-quality face feature, and when the subscript of y is b, it represents the low-quality face feature; λ is a preset coefficient, which can be selected according to actual needs, and is not specifically limited here.

[0113] S105, determining a third correlation degree between the high-quality face feature and the low-quality face feature, and determining a distribution entropy between the high-quality face feature and the low-quality face feature according to the third correlation degree.

[0114] In a specific implementation, a preset face quality distribution entropy evaluator can be used to determine the distribution entropy between the high-quality face feature and the low-quality face feature according to the third correlation degree between the high-quality face feature and the low-quality face feature.

[0115] It should be noted that the preset face quality distribution entropy evaluator can be built by any convolutional neural network.

[0116] Specifically, the distribution entropy between the high-quality face feature and the low-quality face feature can be calculated based on the following formula:

[0117]

[0118] wherein, μ(y g , y b ) represents the distribution entropy between the high-quality face feature and the low-quality face feature; represents the third correlation degree between the high-quality face feature and the low-quality face feature; represents the correlation degree between the pth low-quality face feature and the qth low-quality face feature; k represents the number of normal distribution sampling times; represents the pth high-quality face feature; represents the qth low-quality face feature.

[0119] It should be noted that the third correlation degree and the correlation degree between the pth inferior face feature and the qth inferior face feature can be calculated according to the formula for calculating the first correlation degree in step S103, and details are not repeated here.

[0120] Here, since the distribution difference between the superior face feature and the inferior face feature is large, the third correlation degree between the superior face feature and the inferior face feature is small, and the distribution difference between the two inferior faces is small, the correlation degree between the two inferior faces is large, so the smaller the distribution entropy is, the more accurate the generation of the superior face feature and the inferior face feature is, and vice versa, the larger the distribution entropy is, the lower the extraction accuracy of the superior face feature and the inferior face feature is.

[0121] S106, fusing the variance entropy, the relative entropy and the distribution entropy to generate a training feature value corresponding to the face quality evaluator, and determining a target face quality evaluator that minimizes the training feature value.

[0122] In a specific implementation, after fusing the variance entropy, the relative entropy and the distribution entropy, a training feature value corresponding to the face quality evaluator is obtained, a preset backpropagation optimizer is used to minimize and optimize the training feature value, when the training feature value is optimized to a minimum value, the face quality evaluator training is completed, and at this time, the extraction of the face quality feature and the generation accuracy of the superior face feature and the inferior face feature are all in an optimal state, and the face quality evaluator with the minimum training feature value is used as the target face quality evaluator.

[0123] As a possible implementation, the variance entropy, the relative entropy and the distribution entropy can be fused based on the following method: a corresponding preset weight coefficient is configured for the variance entropy, the relative entropy and the distribution entropy respectively; and the variance entropy, the relative entropy and the distribution entropy are weighted and summed according to the preset weight coefficient to obtain a training feature value corresponding to the face quality evaluator.

[0124] It should be noted that the values of the preset weight coefficients can be configured according to actual needs for the variance entropy, the relative entropy and the distribution entropy, and are not specifically limited here.

[0125] Specifically, the variance entropy, the relative entropy and the distribution entropy can be fused based on the following formula:

[0126] entop = a1(2t-1)var + a2φ + a3μ

[0127] Wherein, entop represents the training eigenvalue; (2t-1)var represents the variance entropy; represents the relative entropy; represents the distribution entropy; represents the preset weight coefficient corresponding to the variance entropy; represents the preset weight coefficient corresponding to the relative entropy; represents the preset weight coefficient corresponding to the distribution entropy.

[0128] In S107, the target variance between the face quality feature and the high-quality face feature is determined by the target face quality evaluator, and based on the target variance and a preset variance threshold, the image quality of the face image to be evaluated is determined.

[0129] In a specific implementation, the target face quality evaluator with the minimum training eigenvalue is used to calculate the target variance between the face quality feature corresponding to the face image to be evaluated and the high-quality face feature generated in the face feature space, and based on the target variance and a preset variance threshold, the image quality of the face image to be evaluated is determined.

[0130] Here, the target face quality evaluator calculates the target variance between the face quality feature and the high-quality face feature in the same way as the calculation of the variance between the face quality feature and the high-quality face feature in the above step S103, and thus no further description is given here.

[0131] It should be noted that the preset variance threshold can be selected according to actual needs, and no specific limitation is given here.

[0132] Specifically, the method for determining the image quality of the face image to be evaluated can be: judging whether the target variance is less than the variance threshold; if yes, determining that the face image to be evaluated is a high-quality image; and if no, determining that the face image to be evaluated is a poor-quality image.

[0133] The method for evaluating face image quality provided in the embodiments of the present disclosure comprises the following steps: obtaining a face image to be evaluated, and extracting a face quality feature corresponding to the face image to be evaluated; sampling a normal distribution vector set conforming to a normal distribution in a face feature space, and generating a high-quality face feature and a poor-quality face feature in the face feature space according to the normal distribution vector set; determining a first correlation degree between the face quality feature and the high-quality face feature, determining a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree by using a preset face quality evaluator; determining a second correlation degree between the face quality feature and the poor-quality face feature, determining a relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature according to the first correlation degree and the second correlation degree; determining a third correlation degree between the high-quality face feature and the poor-quality face feature, and determining a distribution entropy between the high-quality face feature and the poor-quality face feature according to the third correlation degree; generating a training characteristic value corresponding to the face quality evaluator by fusing the variance entropy, the relative entropy and the distribution entropy, and determining a target face quality evaluator that minimizes the training characteristic value; determining a target variance between the face quality feature and the high-quality face feature by using the target face quality evaluator, and determining an image quality of the face image to be evaluated based on the target variance and a preset variance threshold. The face image quality can be accurately judged and distinguished comprehensively.

[0134] Those skilled in the art can understand that, in the above method of the specific implementation, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process, and the specific execution order of each step should be determined according to its function and possible internal logic.

[0135] Based on the same inventive concept, the embodiments of the present disclosure also provide an evaluation device for face image quality corresponding to the method for evaluating face image quality. Since the principle of solving problems in the device of the embodiments of the present disclosure is similar to the method for evaluating face image quality of the above embodiments of the present disclosure, the implementation of the device can be referred to the implementation of the method, and the repeated parts will not be described here.

[0136] Please refer to Figure 3 , Figure 3 The device for evaluating face image quality provided in the embodiments of the present disclosure is shown in the following figure. Figure 3 The device for evaluating face image quality 300 provided in the embodiments of the present disclosure comprises the following modules:

[0137] The feature extraction module 310 is configured to obtain a face image to be evaluated, and extract a face quality feature corresponding to the face image to be evaluated;

[0138] The feature generation module 320 is configured to sample a normal distribution vector set conforming to a normal distribution in a face feature space, and generate a high-quality face feature and a poor-quality face feature in the face feature space according to the normal distribution vector set.

[0139] The variance entropy determination module 330 is configured to determine a first correlation degree between the face quality feature and the high-quality face feature, and determine a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree and by using a preset face quality evaluator.

[0140] The relative entropy determination module 340 is configured to determine a second correlation degree between the face quality feature and the poor-quality face feature, and determine a relative entropy between the face quality feature, the high-quality face feature and the poor-quality face feature according to the first correlation degree and the second correlation degree.

[0141] The distribution entropy determination module 350 is configured to determine a third correlation degree between the high-quality face feature and the poor-quality face feature, and determine a distribution entropy between the high-quality face feature and the poor-quality face feature according to the third correlation degree.

[0142] The training module 360 is configured to generate a training feature value corresponding to the face quality evaluator by fusing the variance entropy, the relative entropy and the distribution entropy, and determine a target face quality evaluator that minimizes the training feature value.

[0143] The image quality evaluation module 370 is configured to determine a target variance between the face quality feature and the high-quality face feature by using the target face quality evaluator, and determine an image quality of the face image to be evaluated based on the target variance and a preset variance threshold.

[0144] The processing flow of each module in the device and the interaction flow between the modules can refer to the related description in the above method embodiments, and will not be described in detail here.

[0145] The device for evaluating face image quality provided by the embodiment of the present disclosure comprises the following steps: obtaining a face image to be evaluated, and extracting a face quality feature corresponding to the face image to be evaluated; sampling a normal distribution vector set conforming to a normal distribution in a face feature space, and generating a high-quality face feature and a poor-quality face feature in the face feature space according to the normal distribution vector set; determining a first correlation degree between the face quality feature and the high-quality face feature, determining a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree by a preset face quality evaluator; determining a second correlation degree between the face quality feature and the poor-quality face feature, determining a relative entropy between the face quality feature and the high-quality face feature and the poor-quality face feature according to the first correlation degree and the second correlation degree; determining a third correlation degree between the high-quality face feature and the poor-quality face feature, and determining a distribution entropy between the high-quality face feature and the poor-quality face feature according to the third correlation degree; generating a training characteristic value corresponding to the face quality evaluator by fusing the variance entropy, the relative entropy and the distribution entropy, and determining a target face quality evaluator that minimizes the training characteristic value; determining a target variance between the face quality feature and the high-quality face feature by the target face quality evaluator, and determining an image quality of the face image to be evaluated based on the target variance and a preset variance threshold. The face image quality can be accurately distinguished and distinguished comprehensively.

[0146] Corresponding to Figure 1 And Figure 2 The face image quality evaluation method in the embodiment of the present disclosure, the present disclosure also provides an electronic device 400, as shown in Figure 4 The structure schematic diagram of the electronic device 400 provided by the embodiment of the present disclosure comprises:

[0147] The processor 41, the memory 42 and the bus 43; the memory 42 is used for storing execution instructions, including the memory 421 and the external memory 422; the memory 421 here is also called the internal memory, which is used for temporarily storing the operation data in the processor 41 and the data exchanged with the external memory 422 such as hard disk, the processor 41 exchanges data with the external memory 422 through the memory 421, when the electronic device 400 runs, the processor 41 and the memory 42 communicate through the bus 43, so that the processor 41 executes Figure 1 And Figure 2 The steps of the face image quality evaluation method in the embodiment of the present disclosure.

[0148] The embodiment of the present disclosure also provides a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program is run by the processor to execute the steps of the face image quality evaluation method described in the above method embodiment. Wherein, the storage medium can be a volatile or non-volatile computer readable storage medium.

[0149] The embodiment of the present disclosure further provides a computer program product, which comprises computer instructions. When the computer instructions are executed by a processor, the steps of the method for evaluating the quality of a face image can be performed. For details, refer to the method embodiments described above, which will not be repeated here.

[0150] The computer program product can be implemented by hardware, software or a combination thereof. In an optional embodiment, the computer program product is embodied as a computer storage medium. In another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (SDK) and the like.

[0151] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the apparatus described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here. In several embodiments provided by the present disclosure, it should be understood that the disclosed apparatus and method can be implemented by other means. The apparatus embodiments described above are only schematic. For example, the division of the units is only a logical function division, and actual implementation can have another division manner. For another example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some communication interface, device or unit, which can be electrical, mechanical or other forms.

[0152] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place or distributed on a plurality of network units. According to actual needs, some or all of the units can be selected to achieve the purpose of the embodiment.

[0153] In addition, each functional unit in each embodiment of the present disclosure can be integrated into one processing unit, or each unit can exist physically, or two or more units can be integrated into one unit.

[0154] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a nonvolatile computer readable storage medium executable by a processor. Based on this understanding, the technical solutions of the present disclosure essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0155] Finally, it should be noted that: the above-described embodiments are only specific embodiments of the present disclosure, used to illustrate the technical solutions of the present disclosure, and not to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art who is familiar with the technical field of the present disclosure can make modifications or easily think of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed by the present disclosure, or make equivalent replacements to some of the technical features. The modifications, changes or replacements do not cause the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A method for evaluating the quality of a human face image, characterized in that, The method comprises the following steps: obtaining a face image to be evaluated, and extracting a face quality feature corresponding to the face image to be evaluated; sampling a normal distribution vector set conforming to a normal distribution in a face feature space, and generating a high-quality face feature and a poor-quality face feature in the face feature space according to the normal distribution vector set; wherein the high-quality face feature is a quality feature corresponding to a face displayed in a face image that is normal in brightness, exposure, and definition and does not have a blur condition; the poor-quality face feature is a quality feature corresponding to a face displayed in a face image that does not satisfy at least one of the following quality evaluation indexes: brightness, exposure, definition, and blur condition; determining a first correlation degree between the face quality feature and the high-quality face feature, and determining a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree by using a preset face quality evaluator; determining a second correlation degree between the face quality feature and the poor-quality face feature, and determining a relative entropy between the face quality feature, the high-quality face feature, and the poor-quality face feature according to the first correlation degree and the second correlation degree; determining a third correlation degree between the high-quality face feature and the poor-quality face feature, and determining a distribution entropy between the high-quality face feature and the poor-quality face feature according to the third correlation degree; fusing the variance entropy, the relative entropy, and the distribution entropy to generate a training characteristic value corresponding to the face quality evaluator, and determining a target face quality evaluator that minimizes the training characteristic value; determining a target variance between the face quality feature and the high-quality face feature by using the target face quality evaluator, and determining an image quality of the face image to be evaluated based on the target variance and a preset variance threshold.

2. The method of claim 1, wherein, The step of sampling a normal distribution vector set conforming to a normal distribution in a face feature space, and generating a high-quality face feature and a poor-quality face feature in the face feature space according to the normal distribution vector set, specifically comprises the following steps: inputting each element in the normal distribution vector set into a preset high-quality face feature generator to determine a high-quality face feature set in the face feature space, wherein the high-quality face feature set is composed of a plurality of high-quality face features; inputting each element in the normal distribution vector set into a preset poor-quality face feature generator to determine a poor-quality face feature set in the face feature space, wherein the poor-quality face feature set is composed of a plurality of poor-quality face features.

3. The method of claim 1, wherein, The step of determining a first correlation degree between the face quality feature and the high-quality face feature, and determining a variance entropy between the face quality feature and the high-quality face feature according to the first correlation degree by using a preset face quality evaluator, specifically comprises the following steps: determining a variance between the face quality feature and the high-quality face feature according to the first correlation degree by using the face quality evaluator; determining a variance entropy between the face quality feature and the high-quality face feature according to the variance.

4. The method of claim 2, wherein, After the determining the second correlation degree between the face quality feature and the poor face feature, and determining the relative entropy between the face quality feature, the good face feature and the poor face feature according to the first correlation degree and the second correlation degree, the method further comprises: Based on the relative entropy, partial differential operation is performed on the good face feature to optimize the accuracy of the good feature generator in generating the good face feature; Based on the relative entropy, partial differential operation is performed on the poor face feature to optimize the accuracy of the poor feature generator in generating the poor face feature.

5. The method of claim 1, wherein, The variance entropy, the relative entropy and the distribution entropy are fused based on the following method: The preset weight coefficients corresponding to the variance entropy, the relative entropy and the distribution entropy are configured respectively; According to the preset weight coefficients, the variance entropy, the relative entropy and the distribution entropy are weighted and summed to obtain the training feature value corresponding to the face quality evaluator.

6. The method of claim 1, wherein, The image quality of the face image to be evaluated is determined based on the target variance and a preset variance threshold, specifically comprising: determining whether the target variance is less than the variance threshold; if less than, determining that the face image to be evaluated is a good image; if greater than, determining that the face image to be evaluated is a poor image.

7. An apparatus for evaluating quality of a face image, characterized by comprising: a face image quality evaluation unit configured to evaluate quality of a face image based on a face image quality evaluation model. Comprise: a feature extraction module for obtaining a face image to be evaluated and extracting a face quality feature corresponding to the face image to be evaluated; a feature generation module for sampling a set of normal distribution vectors conforming to a normal distribution in a face feature space, and generating good face features and poor face features in the face feature space according to the set of normal distribution vectors; wherein the good face features are quality features corresponding to faces displayed in face images with normal brightness, normal exposure, normal clarity and no blurring; the poor face features are quality features corresponding to faces displayed in face images that do not meet at least one of the following quality evaluation indicators: brightness, exposure, clarity and blurring; a variance entropy determination module for determining a first correlation degree between the face quality feature and the good face feature, and determining a variance entropy between the face quality feature and the good face feature according to the first correlation degree by a preset face quality evaluator; a relative entropy determination module for determining a second correlation degree between the face quality feature and the poor face feature, and determining a relative entropy between the face quality feature, the good face feature and the poor face feature according to the first correlation degree and the second correlation degree; a distribution entropy determination module for determining a third correlation degree between the good face feature and the poor face feature, and determining a distribution entropy between the good face feature and the poor face feature according to the third correlation degree; a training module for generating a training feature value corresponding to the face quality evaluator after fusing the variance entropy, the relative entropy and the distribution entropy, and determining a target face quality evaluator that minimizes the training feature value; An image quality evaluation module is configured to determine a target variance between the face quality feature and the high-quality face feature by using the target face quality evaluator, and determine an image quality of the face image to be evaluated based on the target variance and a preset variance threshold.

8. The apparatus of claim 7, wherein, The feature generation module is specifically configured to: input each element in the normal distribution vector set into a preset high-quality face feature generator to determine a high-quality face feature set in the face feature space, wherein the high-quality face feature set is composed of a plurality of high-quality face features; input each element in the normal distribution vector set into a preset poor-quality face feature generator to determine a poor-quality face feature set in the face feature space, wherein the poor-quality face feature set is composed of a plurality of poor-quality face features.

9. An electronic device, comprising: The electronic device comprises: a processor, a memory and a bus, the memory stores machine readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, and the machine readable instructions are executed by the processor to perform the steps of the face image quality evaluation method in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to perform the steps of the face image quality evaluation method in any one of claims 1 to 6.

Images