Research Of Face Recognition Algorithm Based On Block Collaborative Representation

As a kind of biological recognition, face recognition receives great attentionbecause of its friendliness and confidentiality. The classification based on sparserepresentation is widely used in face recognition, through further research we can seeit is collaborative representation improves the accuracy of face recognition. Becausethe collaborative representation classification has the advantages of real-time and highrecognition rate, this paper study the face recognition algorithm deeply based on thecollaborative representation classification.Firstly, the face recognition algorithm based on the regularized least squarescollaborative representation classification using the image itself directly to identify,this will reduce the robustness between the internal characteristics of facial image.Considering the partitioned Gabor feature can reflect more local details of face image,applying block Gabor feature to classification which based on collaborativerepresentation is proposed in this paper. The experiment shows that: the facerecognition based on block Gabor feature collaborative representation classificationimproves the recognition rate further more.Secondly, in real life the small sample size becomes one of the most challengingproblems in face recognition. The face recognition algorithm based on multi-scalepatch with margin distribution optimization and collaborative representationconsidering the different scales of block can supplement the information inclassification, but ignoring the role of the whole image. This paper proposed thatoptimizing the recognition results based on both the whole face image and themulti-scale block, it will improve the stability of the algorithm and slightly increasethe recognition accuracy compared to the original algorithm.Finally, another big difficult problem of face recognition in practical applicationis occlusion. The face recognition algorithm based on block relaxed collaborativerepresentation partitions the occluded image to block, and uses each piece’s intensityas a characteristic. This will lead to losing part information of the face image. And hierarchical multi-scale local binary pattern histogram will be able to extract imageinformation in an all-round way. In order to enhance the robustness of occlusion facerecognition, this paper put forward to jointing the multi-scale local binary pattern tocollaborative representation classification. The experiments based on standard facedatabase show that the proposed algorithm is fast, and having an excellent recognitionrate with the occlusion face image. This will promote the process of face recognitionapplication in real life.