Performance of MPEG-7 edge histogram descriptor in face recognition using Principal Component Analysis

Face recognition is considered as a high dimensionality problem. To handle high dimensionality, a numerous methods have been proposed in literature. In this paper, we propose a novel face recognition method that efficiently solves that problem using MPEG-7 edge histogram descriptor. To the authors´ knowledge, this is the first attempt to use edge histogram descriptor in face recognition. Although MPEG-7 standard represents only local edge histogram we use global and semi-global edge histogram also. We find that local edge histogram mostly helpful for face recognition. We test our system not only using the entire face image as input but also dividing the image into different sub-divisions. PCA is then applied to the edge histogram descriptors of sub-divisions in-stead of raw pixel intensity values of images which traditional methods do. Since we use normalized edge histogram, our face recognition method becomes scale, translation and rotation invariant. Furthermore, our proposed method does not necessarily require all images to be of same resolution as input. We evaluate the proposed method using ORL, Yale and Face94 face databases and achieve superior performance.