A method for improving face recognition accuracy, a computer device and a computer readable storage medium

Abstract

The application provides a method for improving face recognition accuracy, a computer device and a computer readable storage medium, the method comprising: obtaining a face image to be detected and extracting a first face feature; the first face feature is compared with a second face feature in a vector retrieval library respectively according to similarity, and a vector index library is determined according to the similarity, the vector index library comprises names and scores of a plurality of third face features, and the score corresponds to the name; it is judged whether the smallest score is less than a first preset threshold value, if yes, the name corresponding to the smallest score is determined as the name corresponding to the first face feature; if no, it is judged whether the number of times that a target name appears in the vector index library is greater than or equal to a second preset threshold value, if yes, the target name is determined as the name corresponding to the first face feature, if no, the name corresponding to the first face feature is determined according to the number of times that the target name appears and a target score corresponding to the target name, and the method can improve the accuracy of face recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of face recognition, in particular to a method for improving face recognition accuracy, a computer device and a computer readable storage medium. BACKGROUND

[0002] Face recognition technology is a technology that extracts face features based on the current input face image, compares the face features with known face features, and determines the identity of the person in the input face image, which is widely used in attendance, transactions and security fields.

[0003] Face recognition technology mainly has two aspects: (1) based on traditional methods, relying on manually designed features such as HOG and HOF, and machine learning techniques such as support vector machines and principal component analysis; (2) deep learning method, based on convolutional neural network, through a large number of face data samples learning, can achieve a better accuracy. The current mainstream face recognition method is based on deep learning method.

[0004] Face recognition technology based on deep learning method mainly includes: (1) face positioning: positioning the position of face in the image, generally using a rectangular frame to return the coordinates of the face; (2) face key point positioning: representing face through the key features of mouth, eyes, nose hair, eyebrows and face contour; (3) face alignment: through the key points to perform radiation transformation, so that the face is as frontal as possible, which is convenient for better recognition later; (4) face feature extraction: converting the aligned face into a feature vector; (5) face comparison: comparing the feature vector representing the face to get a similarity score, and determining whether two faces correspond to the same subject through the similarity score.

[0005] The existing face recognition technology generally compares the face image to be detected with the face image stored in the vector retrieval library one by one, sets a single threshold, and when the comparison result meets the threshold range, it is considered that the recognition is successful. This method has low face recognition performance in any scene, and has high requirements for the orientation, posture and light of the obtained face image. SUMMARY

[0006] The first object of the present application is to provide a method for improving face recognition accuracy, which can improve the accuracy of face recognition and reduce the quality requirements of face image orientation, posture and light.

[0007] The second object of the present application is to provide a computer device applied to the above-mentioned method for improving face recognition accuracy.

[0008] The third object of the present application is to provide a computer readable storage medium storing the method for improving the accuracy of face recognition.

[0009] To achieve the first object, the present application provides a method for improving the accuracy of face recognition, which comprises the following steps: obtaining a face image to be detected and extracting a first face feature; comparing the first face feature with a second face feature in a vector retrieval library respectively, and determining a plurality of elements of a vector index library according to the similarity, each element of the vector index library comprising a name and a score of a third face feature, and the score corresponding to the name; determining whether the smallest score is smaller than a first preset threshold value, and if yes, determining that the name corresponding to the smallest score is the name corresponding to the first face feature; if the smallest score is not smaller than the first preset threshold value, further determining whether the number of occurrences of a target name in the vector index library is greater than or equal to a second preset threshold value, and if yes, determining that the target name is the name corresponding to the first face feature, and if no, determining the name corresponding to the first face feature according to the number of occurrences of the target name and a target score corresponding to the target name.

[0010] According to the above scheme, the vector index library is determined by comparing the first face feature with the second face feature in the vector retrieval library one by one, so that the name corresponding to the third face feature with the highest similarity can be determined by the first preset threshold value, and when the name cannot be determined by the first threshold value, the name corresponding to the first face feature is determined according to the number of occurrences of the target name in the vector index library and the target score, so that the disadvantage of setting only a single threshold value to identify the first face feature can be avoided, and the third face feature with a lower similarity in the comparison process is considered comprehensively by combining different constraint conditions, so that the accuracy of the final face recognition is improved.

[0011] Further, when the name corresponding to the first face feature is determined according to the target name and the target score, the following steps are included: determining whether the following first determination condition is established: the number of occurrences of the target name in the vector index library is within a first preset range and the average value of the target score is smaller than a third preset threshold value, and if yes, determining that the target name is the name corresponding to the first face feature; if the first determination condition is not established, further determining whether the following second determination condition is established: the number of occurrences of the target name in the vector index library is within a second preset range, and the average value of the first number of items in the target score sorted in ascending order is smaller than the third preset threshold value, and if the second determination condition is established, determining that the target name is the name corresponding to the first face feature; the upper limit value of the second preset range is smaller than the lower limit value of the first preset range.

[0012] Therefore, different constraint ranges and constraint conditions can be determined according to the number of occurrences of the target behavior, so that a multi-threshold recognition mode can be formed, different situations are comprehensively considered, and the accuracy of recognition is improved.

[0013] A further solution is that when the second determination condition is not established, the following steps are performed: determining whether a third determination condition is established: the average of the first values of the target scores is less than a fourth preset threshold value, if yes, determining whether a fourth determination condition is established: the first values of the target scores are the first values of the scores in the vector index library, if the fourth determination condition is established, determining that the target name is the name corresponding to the first face feature, if the fourth determination condition is not established, determining whether a fifth determination condition is established: the second values of the target scores are the second values of the scores in the vector index library, and each of the second values of the target scores is less than a third preset threshold value, if the fifth determination condition is established, determining that the target name is the name corresponding to the first face feature, if the fifth determination condition is not established, determining whether a sixth determination condition is established: the number of occurrences of the target name is greater than a third value, if the sixth determination condition is established, determining that the target name is the name corresponding to the first face feature; the second value is less than the first value; when the third determination condition is not established, determining whether a seventh determination condition is established: the average of the first values of the target scores is less than a fifth preset threshold value, the number of occurrences of the target name is greater than a third value, and the number of different names in the vector index library is greater than or equal to a fourth value, if the seventh determination condition is not established, determining that the target name is the name corresponding to the first face feature.

[0014] Therefore, when the target name does not meet the third preset threshold value, the third preset threshold value can be appropriately relaxed, and other constraint conditions can be added, so that the number of occurrences of the target name can be relaxed, and the name of the face image to be recognized can be recognized when the number of occurrences of the target name is low.

[0015] A further solution is that when the number of occurrences of the target name in the vector index library is within a third preset range, determining whether an eighth determination condition is established: the average of the target scores corresponding to the target name is less than a third preset threshold value, if the eighth determination condition is established, determining that the target name is the name corresponding to the first face feature; the upper limit value of the third preset range is less than the upper limit value of the second preset range.

[0016] A further solution is that before determining whether the number of occurrences of the target name in the vector index library is greater than a second preset threshold value, determining whether the number of different names in the vector index library is greater than a sixth value.

[0017] Therefore, by setting the second preset threshold, it can be quickly determined whether the recognition of the to-be-detected face image meets the condition of continuing the subsequent judgment step, thereby improving the recognition speed.

[0018] A further solution is that, when the first face feature is compared with the second face feature in similarity, and the vector index library is determined according to the similarity, the Euclidean distance between the feature vector of the first face feature and the feature vector of the second face feature is calculated, and each element of the vector index library is determined according to the size of the Euclidean distance.

[0019] Therefore, the Euclidean distance can reflect the absolute difference of individual numerical characteristics, and is suitable for the analysis of reflecting the difference from the size of the dimension, and has higher accuracy than other distances when applied to the recognition method of the present application.

[0020] A further solution is that, before the to-be-detected face image is acquired and the first face feature is extracted, a vector retrieval library is constructed; when the vector retrieval library is constructed, the face data set is collected, and the face data set includes face images of different angles, different postures and different expressions of each person.

[0021] Therefore, the vector retrieval library can be enriched, and the requirements for the angle, posture and expression of the to-be-detected face image can be reduced.

[0022] In order to achieve the second purpose, the present application provides a computer device, which comprises a processor and a memory, wherein the memory stores a computer program, and the computer program is executed by the processor to realize the method for improving the face recognition accuracy.

[0023] In order to achieve the third purpose, the present application provides a computer program, which is executed to realize the method for improving the face recognition accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a flowchart of the method for improving the face recognition accuracy of the present application.

[0025] Figure 2 is a flowchart of the method for improving the face recognition accuracy of the present application.

[0026] Figure 3 is a flowchart of the method for improving the face recognition accuracy of the present application.

[0027] The present application will be further described below in combination with the drawings and embodiments. DETAILED DESCRIPTION

[0028] The present application compares the similarity of the first facial feature and the second facial feature, thereby determining the elements of the patent index library, and then determining the corresponding name of the first facial feature according to the patent index library.

[0029] The method for improving the accuracy of face recognition comprises the following steps:

[0030] The present embodiment is illustrated by recognizing the face image of a public figure in a video and an image, which can be applied to detect whether the face image is used in violation of the rules in online marketing activities, so as to timely protect the rights.

[0031] The present method is realized by a computer program, as shown in Figure 1 The method for improving the accuracy of face recognition comprises the following steps:

[0032] When establishing the vector retrieval library, as shown in Figure 2 Firstly, step S11 is performed to collect the data set. Specifically, a face detection data set and a public figure face data set need to be prepared. The face detection data set is derived from the public WiderFace data set, which has a total of 32203 images, including 393703 labeled faces, 158989 labeled faces for training, and 39496 labeled faces for verification, with a large range of changes in scale, posture, lighting, expression, and occlusion, and the FDDB is used as the test set. The public figure face data set is crawled on the Internet through network crawler technology, and as many multi-angle, posture and expression pictures as possible are selected.

[0033] Then, step S12 is performed to detect the face. The face detection model of the present embodiment takes Retinaface as the main framework, which mainly includes the following 5 modules: backbon-Mobilenet, FPN, SSH, Head, and Multi-task Loss.

[0034] Then, step S13 is performed to determine the face key points. The present embodiment mainly takes the open-source PELD network as the face key point extraction model.

[0035] Then, step S14 is performed to align the face. Specifically, based on the face key points detected in step S13, the five key points of the left and right eyes, the nose tip, and the left and right corners of the mouth are taken out, and the similarity change matrix between the above key points and the pre-defined standard face key points is obtained. Then, affine transformation is performed on the face image to obtain the aligned face.

[0036] Finally, step S15 is performed to extract the facial features. In this embodiment, the MobileFaceNet network structure is mainly used to extract 512-dimensional features from the face. MobileFaceNet is a network structure similar to MobileNetV2, which uses global deep convolution instead of global average pooling layer, and uses PReLU activation layer instead of Relu.

[0037] Based on the above steps, the facial features, i.e., the feature vectors, extracted from the public figure dataset and the names corresponding to the feature vectors can be saved in an array to establish a vector retrieval library. Different index values in the vector retrieval library correspond to a pair of names and feature vectors. The names corresponding to different index values can be the same, for example, the index value 1 in the vector index library corresponds to the name A and the feature vector X, and the index value 17 corresponds to the name A and the feature vector Y, i.e., the name A corresponds to the same person with different facial images, and the expressions and postures of the faces in the images can be different.

[0038] In step S2, the facial image to be detected is obtained and the first facial feature is extracted. Specifically, the feature vector of the first facial feature is extracted in the same manner as the implementation of steps S12 to S15.

[0039] In step S3, referring to Figure 3 , first, step S31 is performed to compare the similarity of the first facial feature with the second facial feature in the vector retrieval library, and to determine a plurality of elements in the vector index library according to the similarity. Specifically, the Euclidean distance, i.e., the score, between the feature vector of the first facial feature and the feature vector of the second facial feature is calculated. In this embodiment, the top 10 scores are arranged from small to large, and the top 10 scores are recorded as D. The feature vectors corresponding to the top 10 scores are third feature vectors, and the index values corresponding to the top 10 scores are recorded as I. Thus, the top 10 scores and the index values corresponding to the scores are stored in the vector index library. According to the index values, the names corresponding to the scores can be determined through the vector retrieval library, i.e., each element in the vector index library includes the name and the score corresponding to the third facial feature value, and the number of different names appearing in the vector index library is counted.

[0040] Then, step S32 is performed to determine whether the smallest score is less than a first preset threshold. In this embodiment, the first preset threshold is 0.5, which is used to represent the threshold of the Euclidean distance required when the matching degree of two facial features is very high. When the smallest score, i.e., the first score D[0][0] in D, is less than 0.5, it indicates that the matching degree of the facial image to be detected and the name corresponding to the score is very high, and the name can be directly determined as the name corresponding to the facial image to be detected. At this time, step S44 is continued to return the name corresponding to the smallest score as the name corresponding to the first facial feature.

[0041] When the minimum score is not less than the first preset threshold, step S33 is performed to determine whether the number of different names in the vector index library is greater than a fifth value. In this embodiment, the fifth value is 8, and the fifth value is used to represent the maximum threshold of the number of different names allowed to appear in the vector index library. In this embodiment, there are 10 scores in the vector index library, and the maximum number of names that can appear in the face image to be detected is 10. When the total number of different names corresponding to the scores exceeds 8, the recognition result of the face image to be detected is not concentrated enough. At this time, step S45 is performed to return a recognition failure.

[0042] When the number of different names in the vector index library is greater than the fifth value, step S34 is continuously performed to determine whether the number of times that the target name appears in the vector index library is greater than or equal to a second preset threshold. The target name is the name with the highest number of appearances in the vector index library. If there are multiple names with the highest number of appearances, for example, among 10 names, Zhang San appears 3 times, Li Si appears 3 times, Wang Wu appears 3 times, and Zhao Liu appears 1 time, the target name is selected from Zhang San, Li Si, and Wang Wu. Specifically, the average values of the three scores corresponding to the names are compared, the name corresponding to the highest average value is the target name, if the average values are equal, the name corresponding to the minimum score among the 9 scores corresponding to the 3 names is the target name, if there are multiple minimum scores and the target names corresponding to the minimum scores are different, the target name cannot be determined, and the recognition is directly returned as a failure. In this embodiment, the second preset threshold is 9, and the second preset threshold is used to represent the minimum number of times that the target name appears when the target name is determined to be recognized successfully only by the number of times that the target name appears. When the number of times that the target name appears is greater than or equal to 9, it means that the face image to be detected is recognized as the same name at least 9 times, then step S47 is continuously performed to determine that the target name is the name corresponding to the first face feature, and then step S48 is performed to return the target name.

[0043] When the result of step S34 is no, it indicates that whether the recognition is successful can not be determined only by the number of occurrences of the target name, and the name corresponding to the first face feature needs to be determined based on the number of occurrences of the target name and the target score corresponding to the target name. At this time, step S35 is continued to be executed to make the first judgment condition, and specifically, the first judgment condition is that the number of occurrences of the target name in the vector index library is within a first preset range, and the average value of the target score is less than a third preset threshold. In the embodiment, the first preset range is a set {7, 8}, and when the number of occurrences of the target name is within the first preset range, it indicates that the face image to be detected at this time has a greater possibility of corresponding to a certain name. At this time, it is necessary to further determine whether the average value of the target score is less than the third preset threshold. The target score is the score corresponding to the target name, and the third preset threshold T = 85. The third preset threshold is a best threshold obtained through experiments on 25,000 face images of non-public figures and more than 10,000 face images of public figures. If the average value of the target score is less than 85, step S47 is continued to be executed.

[0044] If the result of step S35 is no, that is, the first judgment condition is not established, it indicates that the requirement for the number of occurrences of the target name or the target score needs to be further relaxed. At this time, step S36 is executed to determine whether the number of occurrences of the target name in the vector index library is within a second preset range. The second preset range is {4, 5, 6}, and if the number of occurrences of the target name in the vector index library is within the second preset range, it indicates that the face image to be detected has a possibility of corresponding to a certain name. At this time, step S37 is continued to be executed, and otherwise, step S42 is executed. When step S42 is executed, it is determined whether the number of occurrences of the target name in the vector index library is within a third preset range and the average value of the target score is less than the third preset threshold. In the embodiment, the third preset range is {2, 3}, and if the number of occurrences of the target name in the vector index library is within the third preset range, it indicates that the face image to be matched has a smaller possibility of corresponding to a certain name. At this time, if the average value of the target score is less than 85, the eighth judgment condition is established, step S47 is executed, and otherwise, step S43 is executed.

[0045] When step S37 is executed, it is determined whether the average value of the first value of the target score sorted in ascending order is less than the third preset threshold. In the embodiment, the average value of the first 3 values of the target score sorted in ascending order is calculated, and it is determined whether the average value is less than 85. If yes, it indicates that the second judgment condition is established, and step S47 is executed.

[0046] If the result of step S37 is no, i.e. the second judging condition is not established, it indicates that the requirement for the target score needs to be further relaxed, and then step S38 is continued to judge the third judging condition, specifically, the third judging condition is that the average of the first number of items in the target score in the order from small to large is less than a fourth preset threshold. In the embodiment, the fourth preset threshold is 96, and if the average of the first 3 items in the target score in the order from small to large is not less than 96, step S45 is executed. When step S45 is executed, it is judged whether the average of the first number of items in the target score in the order from small to large is less than a fifth preset threshold, and in the embodiment, the fifth preset threshold is 105. When the average of the first 3 items in the target score in the order from small to large is not less than 105, step S43 is executed to return to the recognition failure; and when the average of the first 3 items in the target score in the order from small to large is less than 105, step S46 is continued to be executed. When step S46 is executed, it is judged whether the seventh judging condition is established, i.e. whether the number of times of appearance of the target name is greater than a third number and the number of different names in the vector index library is greater than or equal to a fourth threshold. In the embodiment, when the number of times of appearance of the target name is greater than 4 and the number of different names in the vector index library is greater than or equal to 3, step S47 is continued to be executed, and otherwise, step S43 is executed.

[0047] If the result of step S38 is yes, i.e. the third judging condition is established, step S39 is continued to judge the fourth judging condition, specifically, the fourth judging condition is that the first number of items in the target score in the order from small to large is the first number of items in the score of the vector index library in the order from small to large, i.e. whether the first 3 items in the target score in the order from small to large are one-to-one corresponding to the first 3 items in the score of the vector index library in the order from small to large. If yes, step S47 is executed, and otherwise, step S40 is executed.

[0048] If the result of step S39 is no, i.e. the fourth judging condition is not established, step S40 is executed to judge the fifth judging condition, specifically, the fifth judging condition includes that the second number of items in the target score in the order from small to large is the second number of items in the score of the vector index library in the order from small to large, and each item is less than a first preset threshold. In the embodiment, the second number is 2, i.e. whether the first 2 items in the target score in the order from small to large are the first 2 items in the score of the vector index library in the order from small to large, and the score of each of the first 2 items is less than 85. If yes, step S47 is executed.

[0049] If the result of step S40 is no, i.e., the fifth determining condition is not established, step S41 is executed to determine whether the first second number of items in the target score ranked in ascending order are the first second number of items in the score in the vector index library ranked in ascending order, and the number of occurrences of the target name in the index library is greater than a third value. In this embodiment, the third value is 5, i.e., when the first 2 items in the target score ranked in ascending order are the first 2 items in the score in the vector index library ranked in ascending order, and the number of occurrences of the target name in the index library is greater than 5, i.e., the sixth determining condition is established, step S47 is executed, otherwise, step S43 is executed.

[0050] Computer device embodiment:

[0051] The computer device of this embodiment includes a processor and a memory, the memory stores a computer program, and the processor implements each step of the above method for improving face recognition accuracy when executing the computer program.

[0052] The computer device can include but is not limited to a processor and a memory. Those skilled in the art can understand that the computer device can include more or fewer components, or combine certain components, or different components, for example, the computer device can also include an input / output device, a network access device, a bus, etc.

[0053] For example, the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microcontroller or the processor can also be any conventional processor, etc. The processor is the control center of the computer device, and connects all parts of the computer device through various interfaces and lines.

[0054] The memory can be used to store computer programs and / or modules, and the controller realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory, and calling data stored in the memory. For example, the memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound receiving function, a sound conversion to text function, etc.), etc.; and the data storage area can store data (such as audio data, text data, etc.) created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0055] Embodiment of computer readable storage medium:

[0056] The modules integrated in the computer device of the above embodiment, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the method embodiment for improving face recognition accuracy can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by the controller, the steps of the above method embodiment for improving face recognition accuracy can be realized. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The storage medium can include any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0057] In summary, the method for improving face recognition accuracy of the present application can determine the name of the third face feature corresponding to the highest similarity as the name of the face image to be detected by comparing the first face feature with the second face feature in the vector retrieval library to determine the vector index library, and when the first threshold value cannot be used to determine, the number of times the target name appears in the vector index library, and different constraint conditions are set according to the number of times the target name appears, so that a multi-threshold judgment mode can be formed, avoiding the disadvantages of setting only a single threshold to identify the face image to be detected. By combining different constraint conditions, the third face feature with low similarity in the comparison can be considered comprehensively, thereby improving the accuracy of the final face recognition.

Claims

1. A method for improving face recognition accuracy, characterized in that, The method comprises: obtaining a face image to be detected and extracting a first face feature; performing similarity comparison between the first face feature and a second face feature in a vector retrieval library, and determining a plurality of elements of a vector index library according to the similarity, each element of the vector index library comprising a name of a third face feature and a score corresponding to the name, the score being an Euclidean distance between a feature vector of the first face feature and a feature vector of the second face feature; determining whether the smallest score is less than a first preset threshold, and if so, determining that the name corresponding to the smallest score is the name corresponding to the first face feature; if the smallest score is not less than the first preset threshold, further determining whether the number of occurrences of a target name in the vector index library is greater than or equal to a second preset threshold, and if so, determining that the target name is the name corresponding to the first face feature, and if not, determining the name corresponding to the first face feature according to the number of occurrences of the target name and a target score corresponding to the target name; when the name corresponding to the first face feature is determined according to the target name and the target score, the method comprises the following steps: determining whether the following first determination condition is met: the number of occurrences of the target name in the vector index library is within a first preset range and the average value of the target score is less than a third preset threshold, and if so, determining that the target name is the name corresponding to the first face feature; if the first determination condition is not met, further determining whether the following second determination condition is met: the number of occurrences of the target name in the vector index library is within a second preset range, and the average value of the first number of items in the target score sorted in ascending order is less than a third preset threshold, and if the second determination condition is met, determining that the target name is the name corresponding to the first face feature; the upper limit of the second preset range is less than the lower limit of the first preset range; if the second determination condition is not met, the following steps are performed: determining whether a third judgment condition is met, wherein the third judgment condition is that an average of the items with the first number of values in the target score is less than a fourth preset threshold value, and if so, determining whether a fourth judgment condition is met, wherein the fourth judgment condition is that the items with the first number of values in the target score are the items with the first number of values in the scores in the vector index library, and if the fourth judgment condition is met, determining that the target name is the name corresponding to the first face feature, and if the fourth judgment condition is not met, determining whether a fifth judgment condition is met, wherein the fifth judgment condition is that the items with the second number of values in the target score are the items with the second number of values in the scores in the vector index library, and each of the items with the second number of values in the target score is less than a third preset threshold value, and if the fifth judgment condition is met, determining that the target name is the name corresponding to the first face feature, and if the fifth judgment condition is not met, determining whether a sixth judgment condition is met, wherein the sixth judgment condition is that the target name appears more than a third number of times, and if the sixth judgment condition is met, determining that the target name is the name corresponding to the first face feature; the second number is less than the first number; if the third judgment condition is not met, determining whether a seventh judgment condition is met, wherein the seventh judgment condition is that an average of the items with the first number of values in the target score is less than a fifth preset threshold value, the target name appears more than a third number of times, and the number of different names in the vector index library is greater than or equal to a fourth number, and if the seventh judgment condition is not met, determining that the target name is the name corresponding to the first face feature.

2. The method of claim 1, wherein: if the target name appears a number of times in the vector index library within a third preset range, determining whether an eighth judgment condition is met, wherein the eighth judgment condition is that an average of the target scores corresponding to the target name is less than the third preset threshold value, and if the eighth judgment condition is met, determining that the target name is the name corresponding to the first face feature; an upper limit of the third preset range is less than a lower limit of the second preset range.

3. The method of claim 1 or 2, wherein: before determining whether the target name appears more than a second preset threshold value in the vector index library, determining whether the number of different names in the vector index library is greater than a sixth number.

4. The method of claim 1 or 2, wherein: when comparing the similarity of the first face feature and the second face feature and determining the vector index library according to the similarity, calculating the Euclidean distance between the feature vector of the first face feature and the feature vector of the second face feature, and determining each element of the vector index library according to the size of the Euclidean distance.

5. The method of claim 1 or 2, wherein: Before acquiring the face image to be detected and extracting the first face feature, a vector retrieval library is constructed; When constructing the vector retrieval library, the following steps are included: collecting a face data set, the face data set including face images of each person in different angles, different postures and different expressions. 6.A computer device, comprising a processor and a memory, wherein: The memory stores a computer program, and the computer program is executed by the processor to implement the method for improving face recognition accuracy according to any one of claims 1 to 5. 7.A computer readable storage medium, storing a computer program, wherein: The computer program is executed to implement the method for improving face recognition accuracy according to any one of claims 1 to 5.

Images