Face Recognition Based on Joint Bi-Sparse Representation and Sample Extended Difference Template

In the face recognition, data in each category lie in multiple low-dimensional subspaces of a high-dimensional space respectively. Because the structure information plays a certain support role, we apply the block-structured sparse representation for face recognition. Considering the problem that the training images can not span the facial variation under testing conditions, a novel recognition method of joint bi-sparse representation based sample extended difference template is proposed, which applies an extended difference template to represent the possible variation between the training and testing images. These intra- category variation can be shared by other categories. In other words, the intra- category variation of any category can be represented as the atomic sparse linear combination. So the recognition problem is converted into finding a joint bi-sparse representation of the block-structured sparse representation and atomic sparse representation in the training sample space and extended difference template space. Experimental results on AR and Extended Yale B databases show that the proposed method has better effectiveness and robustness under variable expressions, illuminations and disguises.