The Research And Application Of Face Recognition Based On Local Binary Pattern And Principal Component Analysis

With the rapid development of computer and information technology,face recognition technology has received more and more attention.Interference caused by factors such as illumination,expression,pose and noise will lead to a serious decline in the recognition rate of face recognition algorithms and increase the difficulty of face recognition.Based on consulting a large number of literatures and studying the classic face feature extraction methods such as LBP and PCA,several corresponding improvements were proposed for the shortcomings of the LBP and PCA.Finally,the reliability was verified by experiments.Under the influence of expression and impulsive noise,the face recognition rate is excessively low,therefore a method combining Weighted Information Entropy(IEw)with Adaptive-Threshold Ring Local Binary Pattern(ATRLBP)was proposed(IEwATR-LBP).Firstly,the information entropy was extracted from the sub-block of an original face image,and then IEw of each sub-block was obtained.Secondly,the probability histogram was obtained by using ATRLBP operator to extract the face feature of each sub-block.Finally,the concatenated histogram of original face image was obtained by multiplying IEw with the probability histogram,and the recognition results were calculated through SVM.The results of comparing various algorithms on Yale face database are as follows:the IEwATR-LBP method achieved recognition rates of 98.37%,94.17%,98.20%,and 99.34%on the illumination,pose,expression and occlusion respectively,meanwhile the IEwATR-LBP method achieved the maximum recognition rates of 99.85%on ORL face database.The result of adding noise on ORL was compared among“No-noise”,“Gauss”,and“Salt&Pepper”under average recognition rate of 5 training samples.The rate of“Gauss”was14.04%lower than“No-noise”,but“Salt&Pepper”was only 2.95%lower than“No-noise”.All experimental results show that the IEwATR-LBP method can effectively improve the face recognition rate under the influence of illumination,pose,occlusion,expression,impulse noise.Aiming at addressing the problem that UI-LBP can not occupy a high proportion in face images of all categories,a face recognition method,combining ATRLBP operator and(2D)~2PCA and named(ATRLBP-(2D)~2PCA)briefly,is proposed.Firstly,the ATRLBP operator was used to obtain the face feature from the local face image area,then form a face feature matrix.Secondly,the(2D)~2PCA method was used to extract and reduce the dimension simultaneously of the obtained face feature matrix,moving forward a single step to obtain the result histogram.Finally,using SVM to identify classifications.The results of comparing various algorithms on ORL face database are as follows:the ATRLBP-(2D)~2PCA method obtained 91.1%recognition rate and consumed 3.37s totally.The results of analyzing noise on ORL face database are as follows:when Gaussian noise was added,the ATRLBP-(2D)~2PCA method achieved a recognition rate of 91.0%.Compared with the other four methods,the ATRLBP-(2D)~2PCA method was 1.4,1.6,2.0,and 20 percentage points higher respectively.When salt and pepper noise were added,the ATRLBP-(2D)~2PCA method achieved a recognition rate of 91.0%and was 1.6,1.9,1.3,and 19.7 percentage points higher than the other four methods respectively.The results of comparing various algorithms on Yale face database are as follows:the ATRLBP-(2D)~2PCA method obtained 90.8%recognition rate and consumed 2.75s totally.The results of analyzing noise on Yale face database are as follows:when Gaussian noise was added,the ATRLBP-(2D)~2PCA method achieved a recognition rate of 91.0%.Compared with the other four methods,the ATRLBP-(2D)~2PCA method was 1.5,1.4,3.0,and 22.0 percentage points higher respectively.When salt and pepper noise were added,the ATRLBP-(2D)~2PCA method achieved a recognition rate of 90.0%and was 1.3,1.2,2.0,and 20.7 percentage points higher than the other four methods respectively.All experimental results show that the TRLBP-(2D)~2PCA method dramatically reduces the computational complexity,and achieves the best recognition rate in the two face databases of ORL and Yale respectively.The TRLBP-(2D)~2PCA method has its advantages to varying degrees over time calculation,recognition rate and noise resistance.