Research On The Key Technologies Of Non-complete Face Recognition

With the rapid development of artificial intelligence,.Biometric recognition is becoming a research hotspot,mainly including: voice recognition,fingerprint recognition,and face recognition.Human face is an important symbol of identity information,which has the advantages of uniqueness and not easy to copy for identity verification.Although under certain constraints,face recognition technology has achieved good recognition results,the accuracy of face recognition will decrease because the face images obtained in actual scenes are often incomplete.To solve this problem,this paper proposes two incomplete face recognition algorithms based on the feature extraction of facial organs.1.A non-holonomic face recognition algorithm based on local features of blocks is proposed.First,the face image is logically divided into blocks according to the correlation of face organs,and the neighborhood correlation of local binary patterns(NCLBP)features of each block are extracted.Then,the completeness of each block is judged,and the discriminant results are used to classify and recognize each block locally,so as to reduce the adverse effect of the missing part on the overall recognition result.Finally,the partial recognition results are weighted for voting to obtain the final face category label.Experiments show that this method is aimed at the ORL small sample size set,and has a good recognition rate and recognition speed compared with traditional methods such as SRC and LRC.2.A face recognition algorithm based on non-complete face extraction is proposed.First,the integrity of the extracted NCLBP features of each face block is judged,and the non-integrity sub-blocks are masked.Then,the features of each image block are fused and input into the trained CNN for deep facial feature extraction.Establish a non-complete face recognition model based on segmented NCLBP and CNN(the segmented NLBP and CNN model,TSNCM),and output non-complete face recognition features.Finally,the extracted features are classified using k-nearest neighbor(KNN)to obtain non-complete face category labels.Experiments have proved that this method works when the number of training samples is sufficient.Compared with mainstream CNN-based methods,it has a better recognition rate under the condition of missing faces.In order to verify the effectiveness of the algorithm,a set of incomplete face recognition system was designed and developed.A suitable recognition algorithm can be selected according to the size of the data set,the block feature fusion algorithm is used when the number of samples is sufficient,and the block local feature algorithm is used when the data set is small.The system performance is tested,and the test results show that the developed system has high practicability and recognition accuracy.