Face Recognition Algorithm Of Sparse Representation Based On Extended Dictionary Under The Non-Full Sample Environment

In recent years, because of the superiority, face recognition attract widespread attention. However, face images can easily be influenced by the outside environment. In practical application, it may appear small sample or single sample conditions. To solve this problem, based on the domestic and international relevant research results, this paper will raise some new face recognition algorithms of sparse representation based on extended dictionaryFirst of all, considering the problem that the training images can not span the occlusion variation under testing conditions, we propose a detailed sparse recognition algorithm based on priori sampling constraints and extended dictionary. Priori local sampling sub-modular sparse classify is used to estimate the possible occlusion of testing images. We exploit Borda count for local sub-modular which is unobstructed, thus, the sample information can be obtained. Based on the result of occlusion pattern, global sparse representation are performed to assign votes for the testing images by constructing the sample occlusion extended dictionary. Two classification results are fused to accomplish detailed pattern discrimination in the environment of occlusion.Secondly, to solve the problem of face verification with a single training sample per person, we propose a cost sensitive face verification algorithm based on sparse extended dictionary learning for single sample. The proposed method is adaptive to the gallery set by jointly learning a projection to connect the generic training set with the gallery set. After that, sparse representation classification help us find a training sample which is the most similar to the test sample, then after extracting HOG feature of the training sample and the test sample, calculating the similarity based on the distance criteria is the standard to estimate whether the test sample is within the threshold range; Finally, face robust verification can be accomplished with a single training sample per person under variable illuminations and expressions.Finally, in order to solve the problem of face recognition with a single sample better, the virtual samples extended dictionary sparse representation method for single training sample face recognition based on Borda voted is proposed. Due to the difference of constructing virtual samples, it divides into an extended multiple down-sample Borda voted scheme sparse representation algorithm and an extended adding noise Borda voted scheme sparse representation algorithm, and the recognition results of the two algorithms are compared. The two algorithms achieve satisfactory results under variable illuminations, expressions and pose, especially the former achieves higher recognition rate.