Research On The Face Recognition Method Based On Wavelet Transform And NMFs Algorithm

With the development of some applications such as electronic commerce, face recognition has become one of the most potential biometrics authentication methods. Features extraction has attached importance to people as the most important part of face recognition. The wavelet transformation is a kind of time-frequency signal analysis method. By using it, we can filter out the high frequency information before the feature extraction. And only the low frequency sub-graph is used for recognition. Non-negative Matrix Factorization (NMF) is one of the recently emerged dimensionality reduction methods. Unlike other methods, NMF is based on non-negative constraints. These constraints lead to a parts-based representation because they allow only additive, not subtractive combinations. The non-negative basis vectors can represent the local features of the data. The study of the face recognition method based on wavelet transform and NMFs algorithm has both theoretical and practical values.This paper proposed a feature extraction method by combining Wavelet Transformation (WT) and Non-negative Matrix Factorization with Sparseness constraints (NMFs) together.We firstly give a brief introduction of the evolution of face recognition and NMF algorithm. Various methods of face recognition are presented.Then we discuss the content of NMF algorithm, theoretically analyze the function characteristic and the applying advantage of NMF decomposition method, and introduce the LNMF and NNFs improved from the classical NMF algorithm.The application of Wavelet Transformation to face recognition is discussed, mainly including the definition of wavelet transformation, multi-resolution analysis, two-dimensional separate wavelet transformation and so on.Then, the scheme of human face recognition in this article is proposed and we introduce the experimental process in detail and carry on the analysis to the experimental result. Firstly, we compared the direct NMFs method with the three different wavelet base WT+NMFs methods. Then, the experiments were done under the four kinds of situations that doesn't have cover, have cover in face, eyes, mouth and random area, changing the value of r and the sparseness. Finally, under the two kinds of situations that have cover and doesn't have cover, adopting WT+LNMF method, changing the value of r, we compared the recognition. The experiment showed that this feature extraction method is easy and feasible, reduces the computation complexity, insensitive to the illumination, expression and the part occlusion, has good toughness and high recognition efficiency, especially when the occlusion is serious, the WT+NMFs algorithm has good robustness.Finally we summarize the work and put forward future direction.