Performance Evaluation of Entropy Based LBP for Face Recognition

To improve the accuracy rate of face recognition, a local entropy-based adaptive-weight Local Binary Pattern (LBP) is suggested. A comparative analysis of elliptical and rectangular cropping has been integrated into the methodology for analyzing the recognition rate in different cropping scenarios. Local Binary Pattern is applied on each block to construct local histogram and given as the input K-Nearest Neighbor (K-NN) classifier. A local entropy principle is used here to assign weights to classifier outputs from individual sub-blocks to give more weightage to relevant areas in the face image. Maximum accuracy rates up to 84.1% is achieved in FERET database using rectangular cropping and 76% accuracy is achieved using elliptical cropping with K-NN classifier.