Research On Face Recognition Based On Local Feature Extraction And Deep Learning

Face recognition technology is one of the key technologies in the field of computer vision and artificial intelligence.It plays a pivotal role in the fields of information security,video surveillance,and human-computer interaction.Among them,the extraction of local features of images has been paid attention to by researchers.For global features,local features have better performance in detail description.Starting from the idea of local features and deep learning,this paper explores the face recognition algorithms with better performance under the interference of light,facial expressions and other self or external factors on the basis of existing facial local feature extraction algorithms.The main content and research results of this article are as follows:Aiming at the problem that the features extracted from face images in complex lighting environment are not rich and complete,resulting in low recognition rate,a sparse representation local feature extraction and recognition algorithm is proposed.Based on the traditional CSLBP algorithm,the algorithm first proposes the DTCLBP algorithm,which integrates the information of the center pixel into the CSLBP to dynamically encode each sub-block image,and then uses the CSLDP algorithm to extract the second-order features of the image on this basis,and get the final texture feature.Finally,the sparse representation classifier is used to classify and recognize the extracted features,and the DTCLBP-CSLDP-SRC algorithm is proposed.The algorithm in this chapter does not lose the center pixel information during the two feature extractions,which makes the extracted features more complete and improves the recognition rate of the algorithm.Aiming at the problem that the feature vector extracted by traditional Gabor wavelet transform has high dimensionality and DBN will ignore local information when completing face recognition,a DBN face recognition algorithm based on GCSLBP is proposed.The algorithm first improves the original Gabor transform,introduces the central symmetric local binary mode method to optimize,and then uses the block histogram method to represent the final feature vector,which can better retain image information and reduce dimension.Finally,the deep belief network method is used to improve the robustness of classification and complete the classification and recognition of faces.Aiming at the problem that traditional CNN only extracts features from the entire image and ignores effective local details,a deep convolutional neural network model based on the fusion of global and local features is proposed.First,CNN is divided into two branches,and different network models are used respectively.One is responsible for extracting local features from the image,which can represent more detailed information;the other is responsible for extracting global features from the image,which can express image integrity.Then the local features and the overall features are fused,so that a more discriminative image representation can be obtained,which significantly improves the classification performance.