Recognition Of Special Personnel Based On Integration Of Local And Global Features

Image recognition is a hot issue in the research of computer vision and pattern recognition,which has attracted attention by researchers' more and more in recent years.At the same time,image recognition has broad application prospects in many fields of public safety and daily life,as a biometric identification technology.Special personnel recognition is attracting attention gradually,with the occurrence of violent and terrorist incidents in Xinjiang and the mainland.This paper focuses on the feature extraction problem in image recognition and integrates global and local features to improve the accuracy of special personnel recognition.The main aim of this work is feature extraction of Image recognition,improve the accuracy of special personnel recognition by integrate global and local features.The main works of this research include:1.Verified the recognition of global feature of special personnel.First,obtaining the characteristic subspace of human body image by PCA,calculate feature subspace by linear discriminant analysis,find the eigenvalues and eigenvectors,calculate the optimal classification space.Second,obtaining the characteristic space by PCA and LDA,identifying Characteristics by projection of training sample and test sample.Finally,achieving the purpose of identification by the criterion of nearest neighbor.2.Proposed the method of the gray scale feature of extracting and identifying.First,dividing the body image into N blocks,then get the new grayscale features by mean of gray scale from each subblock,finally verify the precision with BP after PCA.3.Proposed a method based on global and local feature fusion of special personnel,aiming at the problem that the rate of global or local feature recognition is low and easy and susceptible to disturbance.First,extracting the global and local characteristics,and fuse the global and local characteristics,then reduce the new characteristics by PCA,finally,classify the characteristics with BP network,and the recognition rate is higher and more stable.