Face recognition based on LFA features and DLPP algorithm

Some research efforts have shown that face images possibly reside on a nonlinear sub-manifold. Laplacianfaces explicitly considers the manifold structure of the face image. To avoid the singular problem, Laplacianfaces method first project the image vectors to PCA subspaces. However, PCA produces global non-topographic linear filters. In this paper, we propose a novel approach by combining LFA features and DLPP algorithm. LFA can capture local characteristics with little lose of global information and present an effective and robust representation of face images. DLPP avoids the singular problem in LPP, and doesn´t need a PCA transform in advance. Moreover, the new algorithm avoids selecting the subset of LFA features which is difficult and time consuming. A series of experiments show that our algorithm outperforms than Laplacianfaces and has explicit significance.