Face Recognition Based on Multi-Wavelet and Sparse Representation

The feature dimension and redundancy can reduce the face recognition speed and rate, the shading and light changing can heavily affect the face recognition effect. So the first key to face recognition is how to effectively extract face features because the face image contains a lot of redundant information. Multi-wavelet transform has symmetry, orthogonality, compact support and high vanishing moments simultaneously, which can present the face features better than scalar wavelet in each band. By analyzing all the multi-wavelet frequency band components, we can see that the low frequency component of the face image can provide the main feature of a face, and the high frequency components often contain some noises caused by the external interference. Therefore, the low frequency component is often adopted to form the face features. For solving the second problem, we choose the sparse representation recognition method , which has strong robustness to shading and light changing problem. So this paper proposes a face recognition method, which combines multi-wavelet with sparse representation recognition. First, we extract the face features by multi-wavelet, then establish a fully redundant dictionary, and use the sparse representation recognition algorithm for face recognition at last. Based on the YALE, the AR and the FERET face databases, the experimental results show that our method can effectively recognize faces, reduce the dimension of features and has good robustness.