Decision Fusion for Partially Occluded Face Recognition Using Common Vector Approach

摘要

Partial occlusions in the face image negatively affect the performance of a face recognition system. Modular versions of some methods are used to overcome this problem. Modular Common Vector Approach (MCVA) was successfully applied partial occlusion problem. In this work, we apply some well-known decision fusion methods (product rule, borda count, and majority voting) to the decision stage of MCVA approach to increase the classification performance. A well-known appearance based feature descriptor so called Local Binary Patterns (LBP) is used to extract the facial features. The performance comparisons are conducted on AR face database with several experiments. It is observed that combining the classifier outputs using decision fusion methods increase the classification performance of MCVA.