Research On The Human Detection And Face Recognition Based On Image

Computer vision usually means the process of obtaining visual information andpost-processing by computer control and sensor equipment from the surrounding environment,including expression, compression, analysis, processing, storage, and so on, so as to realizethe human biological vision of the function "look". In recent years, the rapid development ofcomputer technology, photovoltaic technology and automation technology, with theemergence of the computer vision system and widely used in the continuous improvement ofscience, industrial, medical and military fields, computer vision technology has become a hotspot of research. Research on computer vision systems has great value and significance bothin theoretical research and practical application research.Human detection tracking and face recognition is the hot research field in computer visionand pattern recognition, it has a broad application prospects in the field of graphics, imageprocessing, intelligent monitoring, smart car system and so on. This paper focuses on somekey issues of human detection tracking and face recognition in image. The details are asfollows:In aspect of moving human detection, human recognition method is widely used in allkinds of videos. But recognition accuracy of these methods is changed negatively because ofcomplexity of background, e.g. void and noise. So a moving body recognition method withbackground robustness is proposed to solve this problem. This method uses tripling temporaldifference method to recognize moving object and dilate the computed binary image withmathematical morphology. And then it uses quadruple directions connection method toconnect disconnection areas. Finally, train the classifier by HOG feature to identify whetherthe moving target is human or not. This method not only needs no restraint of background, butalso shows less computation and higher accurate. Experiment results show its positivecomputational complexity, accuracy and robustness.In aspect of moving human tracking, classic meanshift algorithm is widely used in patternrecognition, but its accuracy decrease large when there is perturbation in background. So tosolve this problem, an improved meanshift method to recognize and track moving body isproposed. This method resets weighting coefficient automatically by detecting tracking area isforeground or not. To compare with exist method, the new method decreases computationamount of background, changed tracking area and predict to moving body. So tracking effectis better than others. Experiment result validates accuracy, effectiveness and robustness.In aspect of illumination pretreatment, illumination pretreatment is an important researchdomain of face recognition, and it has a very important influence to the face recognition, to solve the problem of abnormal light, an improved DOG algorithm to make illuminationpretreatment is proposed. To combine with face reorganization method, compare and analyzelots of characters with experiments of major illumination pretreatment algorithms. Fromexperiment results, the improved DOG algorithm is better than any other ones when there arechanges in illumination and expression of images.In aspect of face recognition, a challenge for face recognition is variation, such as due tolighting or facial expression differences. To solve this problem, this method fusesbidirectional two-dimensional linear discriminant analysis (2DLDA) feature by adaboosttechnique and propose a novel recognition method called AB2DLDA. It can perform wellwith small number of samples.This method uses adaboost to design a classifier and fuses2DLDA and E2DLDA method to improve recognition performance. Experimental resultsshow that the method has good recognition accuracy and robustness.The research based on the moving body detection tracking and face recognition in image isan important subject in a wide range field of computer vision, there are still many difficultpoints which need further research, may this article could give the researchers someinspiration and help.