Research On Improvement Of Distance Criterion Algorithm For Face Recognition

Face recognition technology has been widely applied in recent years,and it is one of the more active research topics in machine learning,pattern recognition and computer vision.Although it has been widely used and has acquired certain effect,there are still many problems to solve,one of the main factors that hinder the development of face recognition is "dimension curse" caused by higher dimensions of the image.Therefore,it is particularly important to use the dimension reduction algorithm to extract the main features of face images.In this paper,we adopt different dimension reduction feature extraction method for dimension reduction,extract the effective and typical features of face image to get relatively few dimension data information,and represent the raw data in the new subspace.Finally,using the classification algorithm to classify the sub-space face data is in order to get the correct data classification.This paper mainly studies dimension reduction algorithms,and proposes two different improved algorithms based on the Maximum Margin Criterion algorithm.First,two-dimensional image dimension reduction methods usually construct objective model using L2 norm or the L1 norm,which have great limitations.To solve this problem and make the algorithm generalize better,a two-dimensional Maximal Margin Criterion algorithm based on L-p norm(1?p?2)is proposed,which improves the adaptability of the original model.Secondly,it is very complicated to solve the two-dimensional Maximum Margin Criterion algorithm based on L1 norm directly,even with a greedy algorithm.The solving process is still complex,and it is difficult to get the global optimal solution,the Particle Swarm Optimization algorithm is not only simple but also can get the global optimal solution.We combine the Particle Swarm Optimization algorithm with the L1 norm-based two-dimensional Maximum Margin Criterion algorithm effectively,we use Particle Swarm Optimization to optimize the dimension reduction projection matrix and utilize the two-dimensional Maximum Margin Criterion of L1 norm to construct the fitness function of particle swarm at the same time in order to make the algorithm to obtain a global optimal solution and ultimately improve the accuracy of face recognition.The validity of the algorithm is verified on classical ORL and Yale databases and additive noise databases.