Bidirectional diagonal Fisher linear discriminant analysis for face recognition
Xu Zhang, Xiangqun Zhang, Yushu Liu · 2009
In this paper, a novel subspace method called Bidirectional diagonal Fisher linear discriminant analysis (BDFLD) is proposed for face recognition. Ensemble classifier is used in order to integrate information of multiple classifiers. BDFLD has the advantage of both 2D FDA and 2D DiaFLD, and directly seeks the optimal projection vectors from diagonal face images without image-to-vector transformation. Also it makes use of two directional diagonal images. The advantage of the BDFLD method over the standard two-dimensional DiaFLD method is, the former seeks optimal projection vectors by interlacing both row and column information of images in two directions while the latter seeks the optimal projection vectors by interlacing both row and column information of images only in one direction. Our test results show that the BDFLD method is superior to standard DiaFLD method and some existing well-known methods.