Multiple representation based sample diversity for face recognition

Proper representation plays a significant role in pattern recognition fields, effective representation of an image makes it easier to be distinguished from images of other classes and can improve the accuracy of classification. In this paper, to obtain appropriate representation of a face, we propose a novel method to represent an original sample. We crop the original image into a number of non-overlapping blocks, and certain operations are applied to each block, then merge all the modified blocks and obtain a new representation of the original sample. Furthermore, we integrate original samples and its corresponding block representations, which provide more local information of images and allow higher accuracy to be achieved. The extensive experiments demonstrate that our proposed method is robust and obtain better accuracy for face recognition in comparison with previous methods.