Increasing the recognition performance in single image per person problem: Combined common feature subspace

Abstract

The recognition performances of many common techniques dealing with face recognition problem mainly depend on the number of images in the training set. However, it is not so easy to collect many different images of one person in realtime applications. In these kinds of applications, training the system with single image can be needed. It is important to have high recognition performances even in these situations. In this paper, a study to increase the recognition performance of Singular Value Decomposition (SVD) based Common Matrix (CM) method, proposed to recognize when the training set has single image, is done. In this study, Combined Common Subspace Method which combines both global and local features of the face is used. The global features are obtained from the whole face image, while local features are obtained from eyes, nose and mouth. The experiments performed on the Ar-Face database show that using the combined common subspace in the SVD based Common Matrix method increases the recognition performance. This increment especially occurs more significantly in the databases containing facial expression differences more significantly. © 2011 IEEE.