Research And Analysis Of Robustness Of Face Recognition Algorithms Based On Sparse Representation Classification

As an important biological feature,face has the characteristic of uniqueness and stability,easy access and not easy to counterfeit and other characteristics,so it is widely used in identity identification,safety monitoring,immigration management,security department photo retrieval and other fields.At the same time,face recognition is an important branch in the field of computer vision.Its research and development not only have theoretical significance,but also have a strong practical value.Sparse representation classification algorithm(SRC)is a new classification algorithm proposed in recent years.It is based on linear representation and sparse constraints,and has excellent classification performance compared to other algorithms.However,in the case of less training samples,with large illumination,attitude,expression and mask,the recognition accuracy of the algorithm declined sharply,so this paper will study from these angles to explore how to improve the robustness of the sparse representation classification.The main research of this paper are as follows:(1)Face recognition based on automatic weighted the K nearest patches for single training sample.When there is only one sample per class,dividing the training samples into blocks,so that the corresponding block composed of sub-block training dictionary to meet the sparse requirements;at the same time with the same rules of test samples are divided into blocks,and each sub block for K neighborhood block,to form a virtual category of test samples.Then dividing the test sample into some sub_blocks,and pick out the K nearest patches of each sub-block to form a virtual testing set and giving automatic weights to those patches in the classification.Finally,using a improved voting mechanism to get classification results of the original test sample..The experimental results show that this method has strong robustness for single sample problem.(2)The sparse representation algorithm based on automatic weighted the blocks by mean square deviation.Although the block processing makes the local information of the face to be preserved,and enhanced the reliability of the algorithm.However,it is necessary to reduce the voting right of the block in the final classification by calculating the mean square error of the block pixel.Compared with the direct block classification algorithm,the algorithm has higher recognition rate.(3)Sparse representation image classification algorithm based on principal component analysis convolution(PCA).The PCA algorithm is used to extract the convolution kernel from the dictionary base,and the image features are extracted by these convolution kernels.Then,weighted and thinned these features,and the sparse representation classification algorithm is used to classify them.Experiments show that the convolution features extracted by this algorithm are more robust and obtain the best recognition effect.