Research On Face Deep Feature Extraction And Clustering

With the rapid development of software and hardware of video surveillance as well as face capture technology,face data has been increasing rapidly.There is an urgent need to achieve the target of analyzing face data by grouping similar faces and extracting valuable knowledge from the face data.In order to accelerate the process,this paper is to extract facial features through the depth learning theory and to group similar faces through the face similarity matrix,which will realize the incremental and fast clustering of large data.The main work and achievements of this paper are as follows.1.The research constructs a face depth feature model.Because of the extensive use of deep learning theory in recent years,the paper adopts the convolution neural network to extract facial features.Firstly,the method constructs the 8 layer network of convolution and uses convolution function to extract the edge feature of the face.Then,the feature dimension reduction is completed by sampling.Finally,the interclass separability and intraclass clustering of deep facial features are verified by experimental analysis.2.In order to solve the problem of dynamic data and large data volume in the process of actual clustering,a fast clustering algorithm combining density peak and representative point analysis is proposed.The algorithm firstly figures out the different features of faces by using the combined Bayesian algorithm to compute the difference between faces,and then dynamically cluster these eigenvalues through the fast face clustering algorithm which combines the density peak and the representative face analysis.The experimental analysis shows that the proposed clustering method can converge rapidly.Then,with the help of the depth feature extraction algorithm mentioned in the previous chapter,the face similarity matrix is constructed to realize the dynamic clustering of face capture incremental data.3.The research develops a system to recognize and analyze faces,and manage face data.Based on the above algorithm,this system is designed for the pharmaceutical industry.It can manage the records of the pharmacy camera and the social security card in cities,and recognize might-be dangerous people through the incremental data of the face data.Main functions of the system include basic data management,photo eigenvalue management,data query and statistics,equipment abnormal monitoring and suspected personnel alarm.