A novel partition selection method for modular face recognition approaches on occlusion problem

Abstract

Recognizing the face with partial occlusion is an important problem for many face recognition applications. Since the occluded parts have no contribution to recognize the face, these parts should be excluded when performing the classification. In this paper, we propose a new method to detect and to use the non-occluded parts of face image for modular face recognition approaches. The occlusion of a partition is decided using the combination of three coefficients which can be easily derived: (i) image entropy, (ii) image correlation, (iii) root-mean-square error. The performance of the proposed partition selection method is tested using the modular extensions of three subspace-based approaches, namely linear regression classification (LRC), common vector approach (CVA), and discriminative common vector approach (DCVA). Modular DCVA is also proposed for the first time in this paper. After the selection of the non-occluded partitions of the face image, LRC, CVA, and DCVA are applied to each of the partitions independently. Then the classifier supports acquired from each of the partitions are combined using three well-known (product, sum, and Borda count) methods to get the final decision. The experiments implemented on the AR and the Extended Yale B face databases show that selection of the face partitions using the proposed strategy improves the recognition accuracy and outperforms state-of-the-art methods.