Research On The Rapid Classification Of Face Recognition Algorithms Based On Collaborative Representation

The face recognition algorithm based on(collaborative representation classification with regularized least squares,CRC_RLS)is an optimized(sparse representation classification face recognition algorithm,SRC)has received extensive attention from scholars.Because it points out that the effectiveness of SCR algorithm classification recognition lies in the synergy of the algorithm itself rather than the sparsity of L1 norm and uses L2 norm instead of L1 norm bypass convex optimization technique to solve the original face recognition problem and make the algorithm execute speed.The algorithm has been improved by 1～3 orders of magnitude,and the algorithm plays an important role in the field of face recognition research.The experiments in this paper show that when the feature dimension of the face image is high,the execution speed of the CRC_RLS algorithm is not ideal,and the performance of the algorithm performance does not meet the requirements of the actual application of face recognition.In order to improve the execution speed of the CRC_RLS algorithm,two sets of new algorithms are proposed in order to improve the execution speed of the algorithm by experiment,observation,analysis and statistics.The experimental results show that the average execution time of the algorithm is shortened from the original algorithm of 2851 milliseconds to 9041 milliseconds to 10.13 milliseconds to 386.5 milliseconds,and the recognition accuracy is the same or very similar.The algorithm in this paper is shortened from 4 steps of the original algorithm to 3 steps.In summary,the algorithm has obvious advantages in terms of execution speed,and the method is simple.The algorithm is applicable to application scenarios with high execution time requirements and small sample size(N<2000).The main contributions of this paper include:(1)We proposed an algorithm model framework based on collaborative coding and image classification.The difference between the original algorithm and the original algorithm lies in the fact that there is no residual process in the new algorithm model framework.This is the root cause of the very short execution time of the algorithm.(2)We proposed two kinds of collaborative classification based fast classification and one-factor face recognition algorithm.The difference between the original algorithm and the original algorithm is that the classification criterion of the new algorithm is based on an element in the coding coefficient vector.The image corresponding to the element affects the classi:fication result.The greatest force,these two algorithms are the basis of the latter algorithm;(3)We proposed two kinds of face vector face algorithm based on collaborative representation fast classification and p-norm fusion coding coefficient.The difference between the original algorithm and the original algorithm is that the classification criterion of the new algorithm is based on the p-norm fusion coding coefficient sub-vector.The image corresponding to this element has the greatest influence on the classification result,which is the reason for the accuracy of the recognition of the two algorithms.