Unsupervised Difference Discriminant Feature Extraction—with Application to Face Recognition

Locality preserving projections (LPP) only concerns the projected locality property while ignores that of "nonlocality". To tackle this problem,a novel unsupervised method of difference discriminant feature extraction is presented. This method extracts an optimal transformation matrix based on maximal nonlocal and local scatter difference,and is successfully applied to face recognition. The proposed method takes into account both the nonlocal and local information to account for the nonlinear structures hidden in the high-dimensional image space. In this method,the "small size sample" problem is avoided by employment of difference operation and the neighborhood relationship is better described by an adequate modification of the adjacency matrix. Extensive experiments on Yale and AR face database demonstrate the effectiveness of the proposed method.