Face Recognition Method Based On Cost-sensitive Hierarchical Extreme Learning Machine

Face recognition is an important branch of biometric identification technology.The granularity information changes of feature extraction in face recognition is a process from rough granularity to precise granularity.In the process,Features will become more representative,decisions will become easier,but the amount of time spent is also increasing.At the same time,face recognition is also a typical cost-sensitive problem.The losses of a unauthorized person was mistakenly allowed to enter the secret service are enormous.In order to solve this problem,image recognition algorithm based on cost-sensitive hierarchical extreme learning machine is proposed in this paper.Hierarchical extreme learning machine extracts features through multiple extreme learning machine sparse auto-encoder and then uses the original extreme learning machine for classification.This algorithm has low computational complexity and high accuracy.This paper analyses that the feature extraction process of hierarchical extreme learning machine is a feature structure of continuous granularity from rough to precise,according to this feature arithmetic combined with sequential three-way decision,the relation between decision cost and time cost can be analyzed by cost-sensitive algorithm,and finally the minimum total cost of face recognition can be obtained.To some extent,this avoids the huge cost of making wrong decisions when there is insufficient information.Sequential three-way decision simulations the human dynamic decision making process,three-way decision is made at each level of the continuous granularity structure.By making three-way decision for each layer of network,the minimum total cost of the algorithm is obtained,it not only retains the advantages of high speed and good generalization of extreme learning machine,but also solves the problem that hierarchical extreme learning machine cannot provide the boundary and insensitive cost.By calculating the total cost and comparing with other traditional deep learning network structure algorithms,experimental results show that the classification algorithm proposed in this paper has better performance of image recognition and less cost.In addition,this paper applies research in cost-sensitive hierarchical extreme learning machine to face recognition tasks and proposes a block face recognition based on cost-sensitive hierarchical extreme learning machine.The face images are divided into proper blocks.Independent sub-blocks can effectively segment the part containing occlusion and make effective information play its role as much as possible.Then,three-way decision is made by building training dictionary in each block and training by the hierarchical extreme learning machine.Experimental results show that the face recognition algorithm proposed in this paper has a better performance on face recognition with noise,and the cost is less.