Multi-mode Identification Of Human Face And Ear Based On Sparse Representation

With the rapid development of information technology,biometric-based personal identification recognition has become a hot topic.Currently,single-mode biometric recognized through face,fingerprint,iris,etc.has been applied in some areas.However,more attention is placed on multimode biometric technology in some high-security applications due to the deficiency of single-mode in noise,non-universality and anti-spoofing capability.This paper is to study multimode fusion recognition on the basis of human faces and ears which are liable to achieve synchronous acquisition.The collection methods are characterized by non-contact,non-invasiveness and easy acceptance.In recent years,with the compressed sensing theory being proposed,biometric recognition technology based on sparse representation theory has attracted numerous scholars at home and abroad.The paper,with the combination of sparse representation theory and multimode biometric recognition,realized multimode fusion recognition algorithm of human faces and ears,which,compared with other biometric recognition algorithms,enjoys high security,excellent recognition performance and low-sensitivity to changes in images.Sparse representation theory is introduced into the paper,for average faces and ears recognition algorithm presents weak robustness to changes regarding image light,facial expression and shooting angle.In the algorithm,PCA,which is able to effectively reduce the sample dimension,is adopted to extract features of human faces and ears.Meantime,feature fusion is used to compress redundant information as well as to distinguish different modal biometric in the largest extent.Given that different modal biometric may be different in contribution to final identification,the fusion algorithm used weighted fusion method.In the proposed algorithm,the test samples' sparse representation coefficient solving in training samples used an iterative faster orthogonal matching pursuit.Experiments proved that theproposed algorithm has better recognition performance and stronger robustness of changes in human faces and ears.Considering the nonlinear separability of training samples and testing samples,the paper further introduced accounting method.After projecting sample feature vectors to high dimensional kernel space which can easily achieve linearly separability,solving the sparse representation coefficients of test sample in over-complete dictionary in kernel space,classified recognition is ultimately achieved by minimizing reconstruction residual.Experiments proved that the proposed nuclear sparse representation recognition algorithm based on human faces and ears further improves recognition performance than its non-nuclear counterpart.