Intelligent analysis attendance checking method and device based on face clustering and file searching by images

Abstract

The invention provides an intelligent analysis attendance checking method and device based on face clustering and file searching through images, and the method comprises the steps: enabling a registration photo and a corresponding site photo to form file data, and carrying out the searching of a to-be-recognized face site photo on a bottom library formed by the face registration photo and the face site photo, the interference caused by large difference between the face registration photo and the to-be-recognized face scene photo is reduced, and meanwhile, through introduction of archive data, coverage of different scene environments can be expanded; through file merging and file accurate rearrangement, the similarity between a to-be-recognized face scene photo and the whole file is fully investigated, so that the correct hit rate is increased; according to the method, archive merging and accurate archive sorting are carried out on the search result, and the similarity between the to-be-recognized face scene photo and all photos in the archives is investigated, so that the reduction amplitude of the recognition correct hit rate is reduced under the condition that the scale of the photos in the base library is relatively large, and a good application effect is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and in particular to an intelligent analysis attendance method and device based on face clustering and image-searching. BACKGROUND

[0002] As a typical application in the field of image recognition, face recognition technology has been implemented on a large scale in security, finance, attendance and other scenarios due to its non-contact and high convenience. Among them, face attendance equipment combines biometric recognition with attendance management, which significantly improves identity verification efficiency and user experience compared with traditional IC card, fingerprint and other attendance methods.

[0003] In the prior art, a static feature library comparison scheme is generally used: by establishing a registration photo database containing employee identity information, the on-site captured photo to be recognized is compared with all registration photos in the database one by one, and the identity information corresponding to the registration photo with the highest similarity is returned as the recognition result. This scheme is essentially a search method based on image similarity, which exposes the following technical defects in actual application:

[0004] Poor anti-interference capability: when the on-site captured photo has light changes, posture shifts, occlusions or low resolution, the visual feature difference between the captured photo and the registration photo will significantly reduce the comparison accuracy;

[0005] Poor cross-scene adaptability: registration photos are usually standard photos taken in a standard environment, while actual captured photos often have large differences due to environmental conditions, leading to feature matching failure;

[0006] Obvious scale bottleneck: as the number of registered personnel increases, the size of the database grows linearly, the false recognition rate rises sharply, and the system response speed also decreases significantly, which seriously affects the use effect.

[0007] The above technical defects restrict the reliability and universality of face attendance systems in high complexity scenarios, so it is urgent to develop a new type of intelligent analysis attendance method to break through the existing technical bottlenecks. SUMMARY

[0008] The present application provides an intelligent analysis attendance method and device based on face clustering and image-searching, which solves the defects of poor anti-interference capability, poor cross-scene adaptability and obvious scale bottleneck in the prior art using a static feature library comparison scheme, and achieves high accuracy and strong adaptability in face recognition attendance.

[0009] The present application provides an intelligent analysis attendance method based on face clustering and image-searching, comprising:

[0010] Calculate the similarity between the to-be-identified face field photo and all photos in the bottom library to determine the most similar candidate photos and the corresponding first similarity score; the bottom library is composed of personnel file data, and the personnel file data includes a personnel face registration photo and several personnel face field photos; each personnel file data has a unique file number, and the personnel face registration photo and the personnel face field photo in the same personnel file data have the same file number as the personnel file data;

[0011] Among all the candidate photos, calculate the mean value of the first similarity scores of the candidate photos with the same file number, and use the mean value as the second similarity score of the personnel file data corresponding to the file number and the to-be-identified face field photo, and construct a nearest file set based on the second similarity score;

[0012] Calculate the third similarity score of the to-be-identified face field photo and each personnel file data in the nearest file set, and sort the personnel file data in the nearest file set according to a preset sorting manner based on the third similarity score, and use the sorting result as the identification result of the to-be-identified face field photo.

[0013] According to the intelligent analysis attendance method based on face clustering and graph search file provided by the application, the construction method of the bottom library is further included:

[0014] Add semi-supervised information to all the face registration photos to keep each face registration photo independent of each other;

[0015] Cluster all the face registration photos and the face field photos through a face clustering algorithm, and cluster the face registration photo and the corresponding several face field photos of each personnel together to form the personnel file data based on the face registration photo;

[0016] Merge the personnel file data of each personnel together to form the bottom library.

[0017] According to the intelligent analysis attendance method based on face clustering and graph search file provided by the application, after using the sorting result as the identification result of the to-be-identified face field photo, the method further includes:

[0018] According to the identification result, the to-be-identified face field photo is stored in the corresponding personnel file data to update the bottom library.

[0019] According to the intelligent analysis attendance method based on face clustering and picture searching for archives provided by the application, the mean value of the first similarity scores of the candidate photos with the same archive number is calculated, the mean value is taken as the second similarity score of the archive data corresponding to the archive number and the to-be-identified face live photo, and the nearest archive set is constructed based on the second similarity score, comprising:

[0020] For N candidate photos {nei i =(id i ,score i ),i∈[0,N)} with the same archive number, the first similarity scores are merged and the mean value is calculated, the mean value is taken as the second similarity score of the archive data corresponding to the archive number and the to-be-identified face live photo, and the nearest archive set is constructed based on The calculation method is as follows:

[0021] c k ∈unique({c(id i ),i∈[0,N)})

[0022]

[0023] M=size(unique({c(id i ),i∈[0,N)}))

[0024] Wherein, score i is the similarity score between the to-be-identified face live photo and the ith candidate photo; id i is the unique number of the ith photo in the database; c(id i ) is the archive number of the photo with the unique number id i ; S ck is the second similarity score of the archive data with the archive number c(id i ) and the to-be-identified face live photo; unique(·) is a de-duplication operation on the candidate photos; mean(·) is a mean value operation on the first similarity of the candidate photos corresponding to each archive data in the nearest archive set; size(·) is the number of elements of the nearest archive set.

[0025] According to the intelligent analysis attendance method based on face clustering and picture searching for archives provided by the application, the mean value of the first similarity scores of the candidate photos with the same archive number is calculated, the mean value is taken as the second similarity score of the archive data corresponding to the archive number and the to-be-identified face live photo, and the nearest archive set is constructed based on the second similarity score, comprising:

[0026] For the nearest archive set The third similarity score of the to-be-identified face live photo and each archive data in the nearest archive set is calculated in the following way: each of the profile data c(id i ) of the third similarity score:

[0027] Update

[0028] wherein s j is the similarity of the to-be-identified face live photo and the photo with a unique number id j in the database; is the third similarity score of the to-be-identified face live photo and the profile data with a profile number c k .

[0029] According to the intelligent analysis attendance method based on face clustering and image-based profile searching provided by the application, before calculating the average of the first similarity scores of the candidate photos with the same profile number, the method further comprises:

[0030] a first weight value is given to all the face registration photos in the database, and a second weight value is given to all the face live photos in the database.

[0031] According to the intelligent analysis attendance method based on face clustering and image-based profile searching provided by the application, the method further comprises:

[0032] The current time and the storage time of each face live photo in the database are obtained, the stored time of each face live photo is calculated, and based on a preset time threshold, the face live photos in the database with a stored time greater than the time threshold are deleted.

[0033] The application further provides an intelligent analysis attendance device based on face clustering and image-based profile searching, comprising:

[0034] A database search module is configured to calculate the similarity of a to-be-identified face live photo and all photos in a database, determine a plurality of candidate photos most similar to the to-be-identified face live photo and corresponding first similarity scores; the database is composed of profile data of each person, and the profile data includes a face registration photo and a plurality of face live photos of a person; each profile data has a unique profile number, and the face registration photo and the face live photos in the same profile data have the same profile number as the profile data;

[0035] A profile set construction module is configured to calculate the average of the first similarity scores of the candidate photos with the same profile number in all the candidate photos, take the average as the second similarity score of the profile data corresponding to the profile number and the to-be-identified face live photo, and construct a nearest profile set based on the second similarity score.

[0036] An archive sorting module is configured to calculate a third similarity score of the to-be-identified face live photo and each of the archive data in the recent archive set, sort the archive data in the recent archive set according to a preset sorting mode based on the third similarity score, and take the sorting result as the identification result of the to-be-identified face live photo.

[0037] The application further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method for intelligent analysis of attendance based on face clustering and graph searching archive when executing the program.

[0038] The application further provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the method for intelligent analysis of attendance based on face clustering and graph searching archive.

[0039] The application provides a method and device for intelligent analysis of attendance based on face clustering and graph searching archive, which calculates the similarity of a to-be-identified face live photo and all photos in a database, determines a plurality of candidate photos most similar to the to-be-identified face live photo and corresponding first similarity scores, calculates the average of the first similarity scores of the candidate photos with the same archive number in all candidate photos, takes the average as the second similarity score of the archive data corresponding to the archive number and the to-be-identified face live photo, constructs a recent archive set based on the second similarity score, calculates the third similarity score of the to-be-identified face live photo and each of the archive data in the recent archive set, sorts the archive data in the recent archive set according to a preset sorting mode based on the third similarity score, and takes the sorting result as the identification result of the to-be-identified face live photo. The application forms archive data by registering a face and corresponding live photos, searches the database formed by the registering photos and live photos by a to-be-identified live photo, greatly reduces the interference caused by large differences between the registering photos and live photos, fully investigates the similarity between the to-be-identified live photo and the entire archive by archive merging and accurate rearrangement of the archive, thereby improving the correct hit rate, and re-investigates the similarity between the to-be-identified live photo and all photos in the archive by archive merging and accurate sorting on the search result, so that the correct hit rate decreases slightly in the case of a large-scale registering photo database, and the application has good application effect. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0041] Figure 1 is a flowchart of the intelligent analysis attendance method based on face clustering and image searching according to the present application;

[0042] Figure 2 is a structural diagram of the intelligent analysis attendance device based on face clustering and image searching according to the present application;

[0043] Figure 3 is a structural diagram of the electronic device according to the present application. DETAILED DESCRIPTION

[0044] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0045] The technical solutions of the present application will be described below in combination with the accompanying drawings. Figure 1 The intelligent analysis attendance method based on face clustering and image searching according to the present application is described.

[0046] As shown in the accompanying drawings, Figure 1 the intelligent analysis attendance method based on face clustering and image searching according to the present application comprises the following steps:

[0047] S1, similarity calculation is performed on the to-be-identified face live photo and all photos in the bottom database to determine multiple candidate photos most similar to the to-be-identified face live photo and corresponding first similarity scores.

[0048] The base library is composed of personnel file data, and the personnel file data includes a face registration photo and several face on-site photos of the personnel, wherein the registration photo has personnel information such as name, membership number, work number, etc., which is pre-registered; the face on-site photo has no personnel information, which is automatically captured by the camera. Each file data has a unique file number c, and the face registration photo and the face on-site photo in the same file data have the same file number as the file data. In addition, all the photos (including face registration photos and face on-site photos) in the base library have a unique number id. For example, the number of a certain photo in the base library can be 56, and the file number can be 9. The number 56 is the identification basis of the photo in the base library, and the file number 9 is the identification basis of the file data to which the photo belongs.

[0049] Specifically, for each photo in the base library, the face on-site photo to be identified captured by the camera in real time is searched in the base library (searched based on face features), and the search result is N photos (i.e. multiple candidate photos) with the closest similarity score (the similarity in the present application represents the similarity of face features, including the first similarity, the second similarity and the third similarity below) to the face on-site photo to be identified in the base library, i.e. N nearest neighbors {nei i = (id i , score i ), i ∈ [0, N]}, wherein score i is the similarity score between the face on-site photo to be identified and the i-th candidate photo in the N photos. In an optional embodiment of the present application, in order to consider the search speed, realize the efficiency and real-time of the system, when the size of the base library is small, a brute force global search can be used; when the size of the base library is large, a fast search algorithm with relatively fast speed can be used, such as quantized hash feature search.

[0050] In an optional embodiment of the present application, the intelligent analysis attendance method based on face clustering and graph searching file further includes a base library construction method, and the construction steps are as follows:

[0051] S11, semi-supervised information is added to all face registration photos, so that the face registration photos are independent of each other.

[0052] Specifically, the purpose of adding semi-supervised information is to identify the independence between all registration photos, which cannot be merged into one file by the clustering algorithm.

[0053] S12, all face registration photos and face on-site photos are clustered by a face clustering algorithm, and based on the face registration photos, the face registration photo and the corresponding several face on-site photos of each personnel are clustered together to form the file data of each personnel.

[0054] Specifically, a face clustering algorithm such as the current DBSCAN algorithm can be used to cluster the face registration photos and the face live photos, the goal being to cluster the face live photos and the face registration photos corresponding to the same person together to form the archive data of the person corresponding to the face registration photo, that is, the archive data of one person is composed of one face registration photo and a plurality of face live photos, and each archive data has a unique archive number c.

[0055] S13, merging the archive data of each person together to form the base library.

[0056] S2, calculating the mean value of the first similarity scores of the candidate photos with the same archive number in all candidate photos, taking the mean value as the second similarity score of the archive data corresponding to the archive number and the face live photo to be identified, and constructing the nearest archive set based on the second similarity score.

[0057] Specifically, for N candidate photos {nei i =(id i ,score i ),i∈[0,N)}, the first similarity scores with the same archive number are merged and the mean value is calculated, the mean value is taken as the second similarity score of the archive data corresponding to the archive number and the face live photo to be identified, and the nearest archive set {(c k ,S ck ),k∈[0,M)} is constructed based on the second similarity score, and the specific calculation method is as follows:

[0058] c k ∈unique({c(id i ),i∈[0,N)})

[0059]

[0060] M=size(unique({c(id i ),i∈[0,N)}))

[0061] Where score i is the similarity score between the face live photo to be identified and the ith candidate photo; id i is the unique number of the ith photo in the base library; c(id i ) is the archive number of the photo with the unique number id i ; c(id iThe first similarity score is calculated by averaging the first similarity scores of the candidate photos in the most recent set of archives. The second similarity score is calculated by averaging the similarity scores of the candidate photos in the most recent set of archives (the second similarity score). This averages the similarity scores of the candidate photos in the most recent set of archives, reducing noise in individual photos and minimizing the impact of outliers. The third similarity score is calculated by averaging the similarity scores of the candidate photos in the most recent set of archives.

[0062] For example, suppose there are 10 candidate photos, uniquely numbered from 1 to 10, with file numbers 1, 3, 2, 1, 1, 3, 1, 2, 1, 2. The similarity scores between these candidate photos and the photo of the face to be identified are 0.9, 0.5, 0.7, 0.8, 0.9, 0.6, 0.9, 0.7, 0.8, and 0.6, respectively. It can be seen that file number 1 includes candidate photos with unique numbers 1, 4, 5, 7, and 9; file number 2 includes candidate photos with unique numbers 3, 8, and 10; and file number 3 includes candidate photos with unique numbers 2 and 6.

[0063] The average similarity score for each group of files is used to obtain the overall similarity score (second similarity score):

[0064] Archive Data 1:

[0065] S1=mean(0.9,0.8,0.9,0.9,0.8)=(0.9+0.8+0.9+0.9+0.8) / 5=0.86;

[0066] Archive Data 2:

[0067] S2=mean(0.7,0.7,0.6)=(0.7+0.7+0.6) / 3≈0.67;

[0068] Archive Data 3:

[0069] S3=mean(0.5,0.6)=(0.5+0.6) / 2=0.55;

[0070] The merged archives yielded duplicate data and their overall similarity scores:

[0071] {(1,0.86),(2,0.67),(3,0.55)}

[0072] It can be seen that the comprehensive matching degree of the to-be-identified face and the archive data 1 is the highest (0.86), followed by the archive data 2 (0.67) and the archive data 3 (0.55) respectively. The number of archives after deduplication (M = 3) is much smaller than the number of original candidate photos (N = 10), which simplifies the subsequent processing steps.

[0073] S3, calculate a third similarity score of the to-be-identified face live photo and each archive data in the nearest archive set, sort the archive data in the nearest archive set according to a preset sorting manner based on the third similarity score, and take the sorting result as the identification result of the to-be-identified face live photo.

[0074] Specifically, for the nearest archive set , the third similarity score of the to-be-identified face live photo and each archive data c (id i ) in the nearest archive set is calculated as follows: j

[0075] Update

[0076] , wherein s j is the similarity of the to-be-identified face live photo and the photo with a unique number id j in the bottom library; is the third similarity score of the to-be-identified face live photo and the archive data with an archive number c k .

[0077] Similarly, taking the nearest archive set {(1, 0.86), (2, 0.67), (3, 0.55)} as an example, all the photos in the archive data C k are extracted again (for example, the archive data 1 contains the photos with numbers 1, 4, 5, 7, and 9), the similarity S j of the to-be-identified face live photo and these photos is recalculated, and the average value is taken as the new similarity k of the archive data C k .

[0078] Suppose after recalculation, the similarity of the to-be-identified face live photo and the photos of the archive data 1, the archive data 2, and the archive data 3 is as follows:

[0079] Archive number Included photo number New similarity (example value) 1 1,4,5,7,9 0.88,0.82,0.85,0.90,0.84 2 3,8,10 0.68,0.70,0.65 3 2,6 0.52,0.58

[0080] The updated recent archive set is {(1, 0.858), (2, 0.677), (3, 0.55)}, that is, the third similarity score of the to-be-recognized face live photo with the archive data 1 in the recent archive set is 0.858, the third similarity score of the to-be-recognized face live photo with the archive data 2 in the recent archive set is 0.677, and the third similarity score of the to-be-recognized face live photo with the archive data 3 in the recent archive set is 0.55. According to the third similarity score, the archive data in the recent archive set is sorted in descending order of the third similarity score, and the sorting result is taken as the recognition result of the to-be-recognized face live photo. Preferably, the TOPK (the first K values in the sorting) in the sorting result is taken as the final recognition result, for example, the TOP1, that is, the archive data 1, is taken as the final recognition result. Through the archive accurate sorting method, the scores of multiple photos in the same archive are combined by using the mean value, the noise of a single photo (for example, the low score 0.52 of the photo numbered 2 in the archive data 3 is evenly distributed) is reduced, the matching degree of the to-be-recognized face photo with the entire archive rather than a single photo can be investigated, which is more in line with the actual scene (for example, multiple photos of a person under different illuminations), and in addition, if a new photo is added to the database, only the corresponding archive of the new photo needs to be recalculated Without modifying the overall process, it is more simple and easy to use.

[0081] In an optional embodiment of the present application, it is considered that the registration photo personnel information is strictly correct (given by a person), and the face live photos in the database are given by a clustering algorithm. In the archive denoising and merging and archive accurate rearrangement links, different database photos can be introduced with a weight value when calculating the similarity of the to-be-recognized live photo and the archive. Specifically, before the step S2 "calculating the mean value of the first similarity scores of the candidate photos with the same archive number", a first weight value is also needed to be given to all face registration photos in the database, and a second weight value is also needed to be given to all face live photos in the database. For example, the face registration photo can be given a weight value λ r , and the face live photo can be given a weight value λ x After the weight values are given, the above formula is updated as follows:

[0082]

[0083] Update

[0084] Wherein

[0085] Update

[0086] Wherein

[0087] Similarly, the above 10 candidate photos are taken as an example for illustration, and it is assumed that λ ris 0.7, λ x is 0.3. The average of the similarity scores of each group of archives is taken to obtain the comprehensive similarity (the second similarity score):

[0088] Archive data 1:

[0089] Suppose that the photo numbered 1 in the archive data 1 is a face registration photo, and the photos numbered 4, 5, 7, and 9 are face live photos.

[0090] S1 = (0.9 x 0.7 + 0.8 x 0.3 + 0.9 x 0.3 + 0.9 x 0.3 + 0.8 x 0.3) / (0.7 + 0.3 x 4) ≈ 0.868;

[0091] Archive data 2:

[0092] Suppose that the photo numbered 3 in the archive data 2 is a face registration photo, and the photos numbered 8 and 10 are face live photos.

[0093] S1 = (0.7 x 0.7 + 0.7 x 0.3 + 0.6 x 0.3) / (0.7 + 0.3 x 2) ≈ 0.677;

[0094] Archive data 3:

[0095] Suppose that the photo numbered 2 in the archive data 3 is a face registration photo, and the photo numbered 6 is a face live photo.

[0096] S1 = (0.5 x 0.7 + 0.6 x 0.7) / (0.7 x 2) ≈ 0.55;

[0097] The archives data after weighting and the comprehensive similarity thereof are obtained:

[0098] {(1, 0.868), (2, 0.677), (3, 0.55)}

[0099] It can be seen that the comprehensive matching degree of the face to be identified with the archive data 1 is the highest (0.868), followed by the archive data 2 (0.677) and the archive data 3 (0.55). By introducing the weight, the similarity calculation is more dependent on the face registration photo, and the recognition robustness can be improved.

[0100] In an optional embodiment of the present application, after taking the ranking result as the identification result of the face live photo to be identified, the face live photo to be identified is also needed to be stored in the corresponding archive data according to the identification result to update the base database, so as to ensure that the base database data gradually becomes strong, and the recognition performance is improved. The updated base database can be directly used as the base database for the next live snapshot photo recognition.

[0101] With the increase of the number of registered personnel, the size of the base library increases linearly, the false recognition rate rises sharply, and the system response speed also decreases significantly, which seriously affects the use effect. Therefore, in order to avoid the influence of the long-time outdated face field photo data in the base library on the response speed, the above-mentioned intelligent analysis attendance method based on face clustering and graph searching file further comprises: obtaining the current time and the storage time of each face field photo in the base library, calculating the storage time of each face field photo, and deleting the face field photo in the base library whose storage time is greater than the time threshold based on the preset time threshold. By deleting the outdated face field photo data, the accuracy of identification is ensured under the condition of ensuring the simplification of the file. For example, assuming that the preset time threshold is 90 days, the storage time of a face field photo in the base library is January 5, 2025, and the face field photo will be deleted after April 5, 2025.

[0102] In summary, the intelligent analysis attendance method based on face clustering and graph searching file provided by the present application determines the most similar multiple candidate photos and the corresponding first similarity score by calculating the similarity of the face field photo to be identified and all photos in the base library; in all candidate photos, the mean value of the first similarity score of the candidate photos with the same file number is calculated, the mean value is taken as the second similarity score of the file data corresponding to the file number and the face field photo to be identified, and the nearest file set is constructed based on the second similarity score; the third similarity score of the face field photo to be identified and each file data in the nearest file set is calculated, and the file data in the nearest file set is sorted according to the preset sorting mode based on the third similarity score, and the sorting result is taken as the identification result of the face field photo to be identified. The present application forms file data by registering photo and corresponding face field photo, and searches the base library formed by the registered photo and the face field photo, which greatly reduces the interference introduced by the large difference between the registered photo and the face field photo; through file merging and accurate file rearrangement, the similarity between the face field photo to be identified and the entire file is fully investigated, so as to improve the correct hit rate; through file merging and accurate file sorting on the search result, the similarity between the face field photo to be identified and all photos in the file is further investigated, so that the correct hit rate decreases little under the condition of large-scale registered photo base library, and the application effect is good.

[0103] Based on the same inventive concept, the present application also provides an intelligent analysis attendance device based on face clustering and graph searching file, which will be described below. The intelligent analysis attendance device based on face clustering and graph searching file described below can be mutually corresponding and referred to with the intelligent analysis attendance method based on face clustering and graph searching file described above.

[0104] AsFigure 2 As shown, the intelligent analysis attendance device based on face clustering and image searching is provided by the application, which comprises a base library searching module 21, an archive set construction module 22 and an archive sorting module 23.

[0105] The base library searching module 21 is used for calculating the similarity between the to-be-identified face live photo and all the photos in the base library, determining the multiple candidate photos most similar to the to-be-identified face live photo and the corresponding first similarity scores; the base library is composed of archive data of each personnel, and the archive data includes a face registration photo and multiple face live photos of the personnel; each archive data has a unique archive number, and the face registration photo and the face live photos in the same archive data have the same archive number as the archive data;

[0106] The archive set construction module 22 is used for calculating the mean value of the first similarity scores of the candidate photos with the same archive number in all the candidate photos, taking the mean value as the second similarity score of the archive data corresponding to the archive number and the to-be-identified face live photo, and constructing the nearest archive set based on the second similarity score.

[0107] The archive sorting module 23 is used for calculating the third similarity score of the to-be-identified face live photo and each archive data in the nearest archive set, sorting the archive data in the nearest archive set according to a preset sorting mode based on the third similarity score, and taking the sorting result as the identification result of the to-be-identified face live photo.

[0108] In an optional embodiment of the application, the intelligent analysis attendance device based on face clustering and image searching further comprises a base library construction module, a base library updating module and a data deletion module.

[0109] The base library construction module is used for adding semi-supervised information to all the face registration photos, so that the face registration photos are independent of each other; clustering all the face registration photos and face live photos through a face clustering algorithm, and clustering the face registration photo and the corresponding multiple face live photos of each personnel together to form the archive data of each personnel based on the face registration photo; and merging the archive data of each personnel together to form the base library.

[0110] The base library updating module is used for storing the to-be-identified face live photo into the corresponding archive data according to the identification result to update the base library.

[0111] The data deletion module is used for obtaining the current time and the storage time of each face live photo in the base library, calculating the storage time of each face live photo, and deleting the face live photo in the base library with the storage time greater than a preset time threshold based on the time threshold.

[0112] Figure 3An example of a schematic diagram of a physical structure of an electronic device is shown in FIG. 1. Figure 3 As shown in FIG. 1, the electronic device can include a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 can communicate with each other through the communications bus 340. The processor 310 can invoke a logical instruction in the memory 330 to execute an intelligent analysis attendance method based on face clustering and graph search provided by each method described above, which includes the following steps:

[0113] Calculating the similarity between the to-be-identified face live photo and all photos in the database to determine a plurality of candidate photos most similar to the to-be-identified face live photo and corresponding first similarity scores; the database is composed of personnel archive data, which includes a personnel face registration photo and a plurality of face live photos; each piece of archive data has a unique archive number, and the face registration photo and the face live photos in the same piece of archive data have the same archive number as the archive data;

[0114] Among all the candidate photos, the average of the first similarity scores of the candidate photos with the same archive number is calculated, and the average is taken as the second similarity score of the archive data corresponding to the archive number and the to-be-identified face live photo, and a nearest archive set is constructed based on the second similarity score;

[0115] The third similarity score of the to-be-identified face live photo and each archive data in the nearest archive set is calculated, and the archive data in the nearest archive set is sorted according to a preset sorting mode based on the third similarity score, and the sorting result is taken as the identification result of the to-be-identified face live photo.

[0116] Moreover, the logical instructions in the memory 330 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts 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 plurality 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 method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0117] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the intelligent analysis attendance method based on face clustering and image searching provided by the above-mentioned methods, which comprises:

[0118] The similarity of the to-be-identified face field photo and all photos in the bottom library is calculated to determine a plurality of candidate photos most similar to the to-be-identified face field photo and corresponding first similarity scores; the bottom library is composed of personnel archive data, and the archive data includes a personnel face registration photo and a plurality of face field photos; each piece of archive data has a unique archive number, and the face registration photo and the face field photo in the same piece of archive data have the same archive number as the archive data;

[0119] In all the candidate photos, the mean value of the first similarity scores of the candidate photos with the same archive number is calculated, and the mean value is taken as the second similarity score of the archive data corresponding to the archive number and the to-be-identified face field photo, and a nearest archive set is constructed based on the second similarity score;

[0120] The third similarity score of the to-be-identified face field photo and each archive data in the nearest archive set is calculated, and the archive data in the nearest archive set is sorted according to a preset sorting mode based on the third similarity score, and the sorting result is taken as the identification result of the to-be-identified face field photo.

[0121] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the intelligent analysis attendance method based on face clustering and image searching provided by the above method, the method comprising:

[0122] performing similarity calculation on the to-be-identified face live photo and all photos in the database to determine a plurality of candidate photos most similar to the to-be-identified face live photo and corresponding first similarity scores; the database is composed of personnel archive data, the personnel archive data includes a face registration photo and a plurality of face live photos of a person, each piece of the personnel archive data has a unique archive number, and the face registration photo and the face live photos in the same piece of the personnel archive data have the same archive number as the personnel archive data;

[0123] calculating the average of the first similarity scores of the candidate photos having the same archive number in all the candidate photos, taking the average as a second similarity score of the archive data corresponding to the archive number and the to-be-identified face live photo, and constructing a nearest archive set based on the second similarity score;

[0124] calculating a third similarity score of the to-be-identified face live photo and each archive data in the nearest archive set, and sorting the archive data in the nearest archive set according to a preset sorting mode based on the third similarity score, taking the sorting result as the identification result of the to-be-identified face live photo.

[0125] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place or distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0126] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software plus necessary universal hardware platforms, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part that makes a contribution to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0127] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features therein can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent analysis attendance method based on face clustering and image-based document search, characterized in that, include: The similarity calculation is performed between the on-site photo of the face to be identified and all photos in the database to determine the multiple candidate photos most similar to the on-site photo of the face to be identified and their corresponding first similarity scores; the database consists of the file data of each person, and the file data includes one registered face photo and several on-site face photos of each person; each file data has a unique file number, and the registered face photo and the on-site face photo in the same file data have the same file number as the file data; Among all the candidate photos, the mean of the first similarity scores of the candidate photos with the same file number is calculated, and the mean is used as the second similarity score between the file data corresponding to the file number and the on-site photo of the face to be identified. Based on the second similarity score, the nearest file set is constructed. Calculate the third similarity score between the on-site photo of the face to be identified and each of the data files in the nearest file set. Based on the third similarity score, sort the data files in the nearest file set according to a preset sorting method, and use the sorting result as the recognition result of the on-site photo of the face to be identified.

2. The intelligent attendance analysis method based on face clustering and image search according to claim 1, characterized in that, It also includes the method for building the base library: Semi-supervised information is added to all the aforementioned face registration photos to ensure that each face registration photo remains independent of the others; The face clustering algorithm is used to cluster all the face registration photos and face scene photos. Based on the face registration photos, the face registration photos and several corresponding face scene photos of each person are clustered together to form the file data of each person. The archive data of each individual is merged together to form the base database.

3. The intelligent analysis attendance method based on face clustering and image-based document search according to claim 1, characterized in that, After using the sorting result as the recognition result of the on-site photo of the face to be identified, the method further includes: Based on the recognition results, the on-site photo of the face to be recognized is entered into the corresponding archive data to update the base database.

4. The intelligent analysis attendance method based on face clustering and image search according to claim 1, characterized in that, Calculate the mean of the first similarity scores of the candidate photos with the same file number, and use the mean as the second similarity score between the file data corresponding to the file number and the on-site photo of the face to be identified. Construct a set of nearest files based on the second similarity score, including: For N candidate photos {nei i =(id) i score i ),i∈[0,N)}, merge the first similarity scores of files with the same file number and calculate the mean, and use the mean as the second similarity score between the file data corresponding to the file number and the on-site photo of the face to be identified, and construct the nearest file set based on the second similarity score. The specific calculation method is as follows: c k ∈unique({c(id i ),i∈[0,N)}) M=size(unique({c(id i ),i∈[0,N)})) Among them, score i The similarity score between the live photo of the face to be identified and the i-th candidate photo; id i c(id) is the unique ID of the i-th photo in the database. i The unique identifier is id. i The file number of the photo; For file number c(id) i The second similarity score between the archive data and the on-site photo of the face to be identified; unique(·) is used to remove duplicates from the candidate photos; mean(·) is used to average the first similarity scores of the candidate photos corresponding to each of the archive data in the nearest archive set; size(·) is used to calculate the number of elements in the nearest archive set.

5. The intelligent analysis attendance method based on face clustering and image-based document search according to claim 4, characterized in that, Calculating the third similarity score between the on-site photo of the face to be identified and each of the data in the most recent archive set includes: For the most recent file set The comparison between the live photo of the face to be identified and the most recent file set is calculated using the following method. Each of the aforementioned archive data c(id) i The third similarity score of ) Among them, s j The on-site photo of the face to be identified is compared with the unique ID in the database. j The similarity of the photos; The on-site photo of the face to be identified and the file number c k The third similarity score of the archive data.

6. The intelligent analysis attendance method based on face clustering and image-based document search according to claim 5, characterized in that, Before calculating the mean of the first similarity scores of the candidate photos with the same file number, the method further includes: A first weight value is assigned to all the registered face photos in the database, and a second weight value is assigned to all the live face photos in the database.

7. The intelligent analysis attendance method based on face clustering and image-based document search according to any one of claims 1 to 6, characterized in that, The method further includes: The system obtains the current time and the storage time of each of the facial scene photos in the database, calculates the storage time of each facial scene photo, and deletes facial scene photos in the database whose storage time is greater than the preset time threshold based on a preset time threshold.

8. An intelligent analysis attendance device based on face clustering and image-based document search, characterized in that, include: The database search module is used to calculate the similarity between the on-site photo of the face to be identified and all photos in the database, and to determine the multiple candidate photos most similar to the on-site photo of the face to be identified and their corresponding first similarity scores; the database consists of the file data of each person, and the file data includes one registered face photo and several on-site face photos of the person; each file data has a unique file number, and the registered face photo and the on-site face photos in the same file data have the same file number as the file data; The file set construction module is used to calculate the average of the first similarity scores of the candidate photos with the same file number among all the candidate photos, use the average as the second similarity score between the file data corresponding to the file number and the on-site photo of the face to be identified, and construct the nearest file set based on the second similarity score; The file sorting module is used to calculate the third similarity score between the on-site photo of the face to be identified and each file data in the nearest file set. Based on the third similarity score, the file data in the nearest file set are sorted according to a preset sorting method, and the sorting result is used as the recognition result of the on-site photo of the face to be identified.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the intelligent analysis attendance method based on face clustering and image search as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intelligent analysis attendance method based on face clustering and image search as described in any one of claims 1 to 7.

Images