Face Recognition Algorithm Based Monogenic Binary Coding And Sparse Coding

Face recognition is an popular yet challenging topic in machine learning and digital image processing areas. Because of its non-touchment and low-expense in system designing, face recognition has a wide range of applications including information security, human-computer interaction, artificial intelligence, and electronic commerce etc. But there are all kinds of uncontrolled environments such as occlusions, which degrade the performance of recognition system. The better recognition system decrease quickly in performance, especially the conditions change harshly. The feature extraction and dimension reduction and classification recognition has been extensively studied in this article. The main works of this dissertation can be summarized as follows:(1) A simple description of face recognition have been made,and the current development status of face recognition have been thoroughly analyzed and introduced.(2)The existing algorithms of holistic and local feature extraction have been summarized, The holistic features are usually very sensitive to the variation of pose and occlusion. It is used to a rough match, while local features generally are used to fine confirmation of face recognition.(3)The face recognition algorithm based on monogenic binary coding has been studied. Efficient face representation is the key for a robust face recognition algorithm. The algorithm of fusing the monogenic amplitude, phase and orientation components is the one of the most effective local feature extraction approach to face recognition.(4)The face recognition algorithm based on sparse representation has been studied. The models of sparse coding are plentiful. The sparse representation based classification and collaborative representation based classification are mainly studied. The sparse representation based classification classifies by minimal reconstruction error. The computational cost of SRC is very high. while collaborative representation based classification is the one of the most effective models in sparse coding.(5)Although the SRC based FR scheme proposed is very creative and effective, there are a issue to be further addressed. The features of Eigenface, Randomface and Fisherface tested are all holistic features. Since in practice the number of training samples is often limited, such holistic features cannot effectively handle the variations of illumination, expression, pose. To this end,a novel face recognition algorithm of fusing monogenic binary coding and sparse coding is proposed; in which the dimensionality problem is resolved by sparse representation. Considering the accuracy and efficiency, we chose collaborative representation based classification. The proposed algorithm is evaluated on benchmark face databases, including ORL, AR and PolyU-NIR. The results indicate a significant increase in the performance when compared with state-of-the-art face recognition methods.