Research On Face Recognition Algorithm Based On Compressed Sensing Theory

Face identification technology is a classic research topic in the field of image processing, with characteristics of low cost, non-violative and easily to operate. Now it has been widely applied in the area of identity authentication and security monitoring. It has great significance on academic research and promoted the rapid development in the area of computer vision, machine learning and etc.Based on great quantity of data samples and theory of utilizing statistical learning, the existing face recognition algorithm is able to analyze and integrate low dimensional data reasonably, however, the poor performance on the process of high dimensional data is not satisfying. The emergence of compression perception theory broke the bottleneck of face recognition. By using of the theory of signal sparse representation, it overcomes difficulty of great amount of sampling data and the challenge of high dimension data.The most classic algorithms of this theory is sparse representation classification algorithm which directly process primary data to prevent information loss from happening in the period of the image preprocessing.In case human face is covered and cannot be 100% displayed, the existing theory will not able to ensure accuracy of signal recovery and consequently the recognition effect. Hence the main research contents in this paper aim to address this issue. Proposing the method of improved sparse representation with segmentation processing. Process covered image and sample database concurrently and identically, and identify each image segment respectively by calculating the residual to get the recognition result. During the process, it need to combine a very sparse random measurement matrix, make use of Sparsity Adaptive Matching Pursuit and SAMP to reconstruct the signal. Under the premise of unknown signal sparsity, system will constantly estimate signal sparsity through iterative in order to solve instability of signal sparsity and achieve the purpose of improving recognition precision and robustness.The design authenticate the suggested algorithm improvement in the AR human facial database and the face of self-built test in library, experimental results show the improved algorithm can effectively remedy the negative influence when covering happens, which is more precise, faster and effective than traditional face recognition method.