Study Of Face Recognition Algorithms Based On Improved Machine Learning

In recent years,face recognition technology has been widely used in network communication,education and security with the development of pattern recognition and machine vision.At present,the face recognition technology has been relatively mature in a limited environment,with the characteristics of high recognition accuracy and fast recognition speed.However,under non-limiting conditions,face images have non-limiting factors such as illumination changes,posture changes,and partial missing facial information.Non-limiting factors can cause problems such as feature loss,alignment error and local aliasing,which seriously affect the recognition effect and face recognition application range.In order to reduce the influence of illumination changes and posture changes on face recognition,the paper proposes face recognition method based on CS-LBP and Deep Believe Network(DBN)for the suppression of illumination and posture changes by CS-LBP.In order to reduce the influence of partial missing information on face recognition,this paper proposes face recognition method based on partitioning CR,because of the ability of CR to separate real face image and partial lack information.Firstly,research on Face Recognition Method based on CS-LBP and DBN(FRMCD).On the one hand,the CS-LBP operator which can describe the local micro-pattern of the image has the characteristics of suppressing the brightness change of the pixel,reducing the coding time and the anti-noise ability.On the other hand,the DBN has the characteristics of automatic learning image abstract information and the active factor intervention.Therefore,the paper proposes the face recognition method that applies CS-LBP face image local information and uses DBN for learning classification.On this basis,a large number of face image recognition experiments and the comparison experiments with existing methods were performed to verify the FRMCD effectiveness on three common facelibraries with posture changes,illumination changes,and image occlusion),namely ORL face database,Extend Yale B face database and CMU-PIE face database.The results show that when FRMCD is applied to face recognition with illumination changes and posture changes,it has a good recognition effect;when the training samples are insufficient,the advantage of FRMCD is more significant;Compared with the FRMLD,the time consumption is greatly reduced.Secondly,research on Face Recognition Method based on partitioning CR(FRAPCR).The missing information in face images is uncertain.This paper uses CR to separate it from the original image,which can effectively reduce the influence of face recognition results of missing information maps.The separated images are segmented,the sparse coefficients of each sub-block are obtained to classify,and the final category labels are completed statistically.The determination can effectively reduce the influence of the invalid classification with the smallest sparse error obtained by one or more sub-block features on the whole image.On this basis,using the ORL face database,Extend Yale B face database and AR face database with occlusion images,when the face image has some missing information(pixel information missing,corrupted block and occlusion),comparison experiments of FRAPCR with existing face recognition methods were completed.The experimental results show that FRAPCR has high recognition rate and stable recognition rate when the partial image information missed.