A robust two quadrant sparse classifier for partially occluded face image recognition

In the last decade, Sparse representation method of classification has proved its efficiency for face recognition applications. The major challenge for face recognition algorithms is partial occlusion and expression changes in face images. In case of upper or lower face occlusion, a large portion of discriminative information gets missing which results into lower classification accuracy. In this paper a novel two quadrant sparse classifier is proposed which implements sparse representation in two different quadrants. Proposed classifier uses visual face information present in upper and lower quadrants independently to classify the test face image which provides correct classification even if one quadrant is occluded. The enhanced accuracy of proposed technique is proved using extensive simulations carried out on two standard databases (ORL and YALE). The experimental results for all number of training images and training sets are compared with simple sparse method in terms of mean classification accuracy. The performance of proposed technique is analysed with a maximum improvement of 3.51 % in terms of classification accuracy.