Research On Robust Face Recognition Based On Sparse Representation

In recent years,face recognition has become a hot research field of pattern recognition and artificial intelligence.Due to the facts of the convenient information collection and the non-contact,the face recognition technology has been widely applied to the fields of security monitoring and information management.Feature selection is a very important step in face recognition,and texture feature is one of the important features for the image.Because of its good adaptability to the changes of illumination,shifting,angle,texture feature has been more and more applied to face image classification.A large number of studies show that an effective feature extraction method can not only make the feature vector to carry more useful information,and improve the recognition rate and robustness,also can reduce the scale of data,increase the calculation speed.At present,the technology of face recognition which based on sparse representation algorithm,for its simple model and better recognition effect,has attracted a lot of researchers.In order to improve the robustness and recognition rate of the system,the paper will apply feature extraction to construct the dictionary set of sparse representation algorithm.The main contributions are described as follows:(1)Through researching the compressive sensing and sparse representation theory,we know that the sparse coefficient has a strong description for the signal class.It can select the closest category which can represent the input vector,and ignore the category that is not closely related.The simulation experiments on AR face databases show that this method is feasible to recognize the face.(2)Through researching the Local Color Vector Binary Patterns(LCVBP),the color norm and angular patterns,which are extracted from the multi-signal channel characteristics of color images,have a higher feature dimension and greater computational cost.Hence,a novel and automatic feature extraction approach is presented for face recognition based the LCVBP method.On the basis of YCbCr color space,the feature points in a face image,such as eyes,nose and mouth,are located,and the feature regions are obtained by utilizing the location of feature points.Finally,the LCVBP histograms of these feature regions are extracted and put together in sequence as the final histogram characteristics of an image.The experimental results show that the presented method can obtain the approximately equal identification rate with the LCVBP method by abandoning this redundant information in a face image.However,the presented method has lower dimension of characteristic vector,smaller calculation cost,and faster face recognition speed.(3)In order to improve the robustness of sparse representation algorithm,an input mage authenticity discrimination technology is introduced based on the original algorithm,which can effectively identify the non-library face.By introducing the color image information into the sparse representation model,and combining with the local color vector binary mode to extract feature,an improved non-uniform fusion of multiple features of the feature extraction method is proposed to construct dictionary training.Through adding the error term to the original algorithm,we are able to reduce the occlusion effect on recognition rate effectively.By finishing several comparative experiments in the color FERET face image database and the color face database of California Science Institute respectively,the obtained simulation results show that the proposed method can obtain better recognition rate,and have a better recognition effect on the faces which have rotation about 15 degrees.