Face Recognition Algorithm Based On Sparse Representation

As an important biometric technology, face recognition extracts visual characteristic information from facial images to achieve identification and has very broad prospects for development. First, face recognition is a secretive non-contact operation with uniqueness and reliability, so it can be widely applied in many fields such as video surveillance and ID authentication. Second, the research of face recognition has great theoretical value, for it has involved many overlapping academic discipline such as digital image processing, pattern recognition, machine learning and so on. The narrow sense of face recognition includes two parts:feature extraction and classifier design. Feature extraction is used to reduce dimension and extract useful information for classification, and the classifier design part is to correctly classify the features that extracted.The Sparse Representation Theory is a hot research topic of face recognition field in recent years. The theory indicated that a given sample can achieve the best sparse reconstruction with all the other samples in the database. Under ideal condition, it is only associated with the samples from the same class. So we can use the sparse coefficients to characterize the relationship between samples, which is of great significance to the face recognition area. On the other hand, graph embedding model provides a unified research framework for the current feature extraction algorithms, and they are just different in objective function and algorithm constraint. Therefore, we can use graph embedding to develop new models of feature extraction algorithms.This paper firstly proposed a new feature extraction algorithm-Sparse Representation Discrimination Analysis (SRDA) by combining the sparse representation theory and the graph embedding model together. SRDA algorithm has good reorganization performance, for it can maintain not only the sparse reconstruction relationship of original data, but also the spatial structure in low dimensional space. Then we combined the SRDA algorithm with Gabor features, namely, Gabor-based SRDA algorithm (GSRDA), which can enhance the recognition rate. In addition, the paper first improved the Sparse Representation Classifier, using sum of K largest sparse coefficients each type for classification, and also applied the SVM method with error correction ability as a classifier. Together with the two classifiers, the latest feature extraction algorithm achieved the best recognition performance, which is proved by experiments on ORL, AR and FERET database.