Feature Extraction And Classification For Face Recognition

Face recognition is one of the hot topics in the field of pattern recognition and computer vision.Face recognition is a typical problem of image analysis,understanding and classification computing.Face recognition propose a good issue for pattern recognition,computer vision and Brain Science and Brain-Like Intelligenc.They promote each other.Face recognition does not influence the work of the probe and could be easily implemented.In particular,the problem of the social public security is very serve.Face recognition could be concealed aunique advantages and has great value in identification,intelligent retrieval,security survelliance,and financial and other fields.We know that feature extraction and classification are ones of the most basic and the most important problems in the field of pattern recognition.Extracting the effective features or key features is a shortcut for enhancing the recognition accuracy.Feature extraction contains:feature generation,feature selection and dimension reduction.Since there are only small numbers of high-dimensional training samples,how to extract the key features for dimensionality reduction and recognition is a hot topic with difficulties in current research in the image-based object recognition,particularly,in face recognition.Classification is to how construct a suitale fuction to classify the data into the correct class while not knowing the distribution of the data.The paper mainly focous on face recognition and the main work includes:(1)For feature generation,two local gabor texture feature descriptors methods are proposed for face recognition.(2)For feature extraction,based on collaboration representation based classification,a serious of feature extraction methods are given:collaborative representation analysis(CRA),a nonlinear kernel collaborative representation based classification(KCRC),symmetrical collaborative representation analysis(SCRA)and collaborative representation reconstruction based projections(CRRP).(3)For classification,a local mean representation based classifier(LMRC)and tis kernel version are proposed.We build the dictionary using the local class means and sparsely linear reconstruct the probe.LMRC is easy to be solved with a least squares.(4)Based on the nuclear norm-based matrix regression,matrix regression is performed on each patch and the outputs from all patches are fused for face recognition with occlusion and illumination variants.Then we give a multi-scale patch-based matrix regression scheme for face recognition.