Effective discriminant feature extraction framework for face recognition
Yan Yan, Yu‐Jin Zhang · 2010
It is well known that extracting effective features from images is a crucial step for appearance-based face recognition methods. In this paper, an effective framework for extracting discriminant features, by so called Discriminant Class-dependence Feature Analysis (DCFA), which combines Linear Discriminant Analysis (LDA) and 1-D Class-dependence Feature Analysis (1D-CFA), is proposed. From one side, LDA extracts features to discriminate all classes while it cannot distinguish close classes well. On the other side, 1D-CFA extracts features to emphasize one specific class and suppress other classes. By taking advantages of the merits of two different and complementary feature extraction methods, DCFA can extract discriminant features very effectively. Except the analysis, the experimental results on three well-known face recognition databases also demonstrate the effectiveness and robustness of the proposed approach.