基于多尺度Retinex算法结合PCA特征加权的人脸识别方法

Face Recognition Method Based on Multi-scale Retinex Algorithm Combined with the Feature Weighting of PCA

结合多尺度Retinex算法和PCA算法的特点,并引入权重系数,提出了一种新的人脸识别方法.首先,在人脸图像的预处理阶段,利用多尺度Retinex算法提取人脸图像光照不变分量,然后用PCA算法提取人脸光照不变量的主特征;为进一步减少光照变化对人脸识别率的影响,对提取到的主特征的前两个向量加小于1的权重系数;接着利用k近邻分类器进行人脸分类识别;最后基于CAS_PEAL_R1光照子集人脸库,在Matlab环境下进行仿真实验,实验结果表明该方法提高了人脸识别率.

The face recognition rate is not high under Lighting-variation conditions,combining with the characteristics of multi-scale Retinex algorithm and PCA algorithm in this paper,to further introduce the weight coefficient of processing,a new method of face recognition is presented.First of all,extract the Lighting-invariant component of the face image with the multi-scale Retinex algorithm in the preprocessing stage of face image,then extract the main features of the face image with the PCA algorithm.In order to further reduce the influence of Lighting-variation on face recognition rate,to plus weight coefficient which less than one for the first two main feature vectors.Face classification by using the K-Nearest Neighbor classifier next.Finally,to program under the environment of Matlab for simulation experiment based on the CAS_PEAL_R1face database,the experimental results show that the method is effective and the face recognition rate is improved.