Faces Recognition with Image Feature Weights and Least Mean Square Learning Approach

Most of 2DPCA-enhanced approaches improve face recognition rate while at the expense of computation load. In this paper, an approach is proposed to greatly improve face recognition rate with slightly increased computation load. In this approach, the 2DPCA is applied against a face image to extract important image features for selection. A weight is then assigned to each of selected image features according to the feature’s importance to face recognition. The least mean square (LMS) algorithm is further applied to optimize the feature weights based on the recognition error rate during learning process in order to improve face recognition performance. The experiments have been conducted against ORL face image database to make performance comparisons among several better-known approaches, and the experimental results have demonstrated that the proposed approach not only has excellent face recognition rate of 99% but also requires only slightly higher computation load than 2DPCA, making the approach more practical to real face recognition applications..