自适应中心对称局部三值模式的人脸识别

摘要

针对局部三值模式描述人脸图像纹理特征时直方图维数过高以及阈值不能自适应选取的缺陷，提出一种自适应中心对称局部三值模式方法。首先，用具有降低维数的中心对称局部三值模式算子对人脸图像编码，把邻域像素均值引入编码中以增强抗噪性能；其次，嵌入统计邻域均值与邻域像素的标准差作为阈值以自适应提取人脸特征，并统计特征直方图；最后用卡方距离度量训练样本特征直方图和测试样本特征直方图的相似度，采用最近邻分类器分类识别。所提算法在 YALE、Extended Yale B 人脸图像库上的最高正确识别率分别达到99．67％，99．33％；识别一张人脸的速度分别达到0．1984和0．3988 s。实验结果表明，所提算法对光照变化和噪声更加鲁棒，有效提高了人脸识别的精度和速度。%In this paper we propose a new method called centrosymmetric local ternary pattern with adaptive threshold (CS-LTPAT)to ad-dress the shortcomings of local ternary pattern in too high the histogram dimension and not being able to adaptively select threshold when de-scribing the texture features of face image.First,the method encodes the face image with the dimension-lowered centrosymmetric local ternary pattern (CS-LTP)operator,and introduces the neighbourhood pixel mean value to encoding for enhancing the anti-noise performance;Second-ly,it embeds the standard deviation of statistical neighbourhood average and neighbourhood surrounding pixel as the threshold to extract the fa-cial feature adaptively,and counts the features histograms.Finally,it uses chi-square to measure the similarity of training sample features histo-gram and test sample features histogram,and employs the nearest neighbour classifier in recognition.The proposed approach is applied to YALE and Extended Yale B standard face database,result shows that the highest correct recognition rates reach 99.67% and 99.33% respec-tively,and the speed of identifying a face reach 0.1984s and 0.3988 s respectively.Experimental result demonstrates that the proposed method effectively improves the accuracy and speed of the face recognition,and is more robust on the illumination variation and noise.