改进的Gabor-CRC人脸识别算法

Improved face recognition algorithm based on Gabor feature and collaborative representation

针对CRC_RLS对光照变化、表情变化、姿态偏转的人脸识别鲁棒性不高和Gabor-CRC算法对有伪装的人脸识别识别率下降的问题,提出一种基于分块Gabor特征和加权协同表示的人脸识别算法(BG-WCRC)。对图像进行分块,分别对每块提取Gabor特征、下采样和PCA降维构成子块Gabor特征字典,利用CRC_RLS方法计算每块的类别,对各块的类别进行加权投票得到图像的最终类别。在AR、Extended Yale B和ORL人脸库上进行实验,实验结果表明,该算法具有较强的鲁棒性。

Aiming at the problems that collaborative representation has low robustness to illumination change,expression change and pose rotation,and that Gabor feature and collaborative representation has low recognition rate facing disguise,a face recognition method was proposed based on Gabor feature by blocks and weighted collaborative representation.The image was divided into some blocks and the Gabor features were extracted for each block.The dimensions were reduced through principal component analysis and Gabor dictionary for the Gabor feature of each block down-sample was structured.The collaborative representation was used to compute the category of each block.The final category of the image was voted by weight for categories of all blocks.The performance of the algorithm was verified on AR,Extended Yale B and ORL databases.Experimental results show that this algorithm has strong robustness.