The Research Of Face Recognition Based On Local Feature And Collaborative Sparse Representation

Face recognition technology is a very hot issue in the field of computer vision and pattern recognition it has been widely used in public security and house monitoring safety,etc. For a long time, there are so many excellent face recognition measures had been formed such as geometric feature model,elastic graph matching model,support vector machines and adaboost,etc. To address the shortages of those techniques for example:the high computational complexity, incomprehensive feature extraction and the low robustness to illumination an pose,etc. This thesis based on compressive sensing theory and sparse representation theory which are developed in recent years, contrast with the traditional methods to stress the superiority of the new theory, summary and analysis the sparse representation use the overall features to classified the characteristics that very sensitive for Illumination changes,local changes and pose changes,etc. At the same time, the thesis will discuss about the collaborative sparse representation classification.Based on the summary and analysis of sparse representation and collaborative sparse representation, the thesis will use the system process as the sequence to start the work. The main task as followers.(1) Introduce the research status of face recognition technology at home and abroad. Brief introduce common face database and the evaluation ruler. With the help of system process the face recognition technology was introduced.(2)Presentation for pretreatment stage and normalization process of human faces, putting forward a eyes and mouth positioning method of combining Sobel operator and Susan operator with shade of gray distribution, processing zoom, editor and rotation of images through normalization basis of position.(3)Presentation for two better methods of local texture feature descriptors are Gabor wavelet and Local binary patterns.Raised improvement LBP operator, decreasing the extracted features overall dimension using consistency operator, improving the shortcoming of primitive operators central pixel easy for ignoring local feature, increasing extracted feature information and decreasing complicated level of feature. The test shows that ULBP operator has better capable of feature description than Gabor wavelet and LBP operator by comparing the extracted feature with Gabor wavelet and LBP feature in human faces library.(4)The collaborative sparse representation classification replace The L1 norm which is used by The sparse representation classification with the minimum least square performs as well as The sparse representation.using Gabor wavelet and Uniform LBP features, the algorithm classify their types with related sparse dictionary. With the help of local texture feature,system recognition rate increases.Transform dictionary by using the ULBP algorithm.discuss and analyze the effects of each feature extract algorithm to collaborative sparse representation algorithm, then the experiments through the face database show that it is effective for the collaborative sparse representation algorithm.