A novel model for Enhanced Principal Component Analysis

摘要

In this paper, a novel mathematical model for Enhanced Principal Component Analysis (EPCA) is proposed. With the new mathematical model, the performance of EPCA could be enhanced in pattern recognition area. Compared with PCA, EPCA could adaptively distinguish different variables of sample vector according to their scale in statistics. The optimization problem of EPCA could be solved in the framework used to solve the optimization problem of PCA, so EPCA dose not require more computational complexity than other improved PCA algorithms. When applied to face recognition, EPCA are robust to different facial expression, different illumination intensity and large variation in lighting direction. EPCA outperforms many famous algorithms (PCA, FLD and ICA) in the experiments on Harvard face database.