基于Gabor特征与加权协同表示的人脸识别算法

Face recognition algorithm based on Gabor feature and weighted collaborative-based classification representation

摘要:

针对基于稀疏表示分类方法的训练样本于与类别标签信息提取不足,特别是在训练样本和待测样本都受到噪声污染的情况下将会明显下降及算法复杂度较高的问题,提出以Gabor特征以及加权协同为基础的人脸识别算法。最初需要对人脸图像内所包含的各个尺度以及方向的Gabor特征完成提取,在稀疏表示中引入Gabor特征,将降维后的Gabor特征矩阵作为超完备字典,再用稀疏表示增强加权协同表示得到该字典下的的稀疏表示系数,然后利用增强系数与训练样本的标签矩阵完成对测试样本进行分类识别,从而得到Gabor特征以及加权的协同表示分类方法,在Yale人脸数据库、Extended Yale B和AR人脸数据库上以及在FERET人脸数据库对人脸姿态变化的实验表明新算法具有更好的识别率和较短的计算时间。

Abstract:

In order to solve the problem of insufficient information extraction of training samples and category labels based on sparse representation classification method, especially when the training samples and samples to be tested are polluted by noise, and the complexity of the algorithm is high, a face recognition algorithm based on Gabor feature and weighted collaboration is proposed. At first, it is necessary to complete the extraction of Gabor features of each scale and direction contained in the face image. Gabor features are introduced into the sparse representation. The Gabor feature matrix after dimension reduction is taken as a super-complete dictionary, and then the sparse representation is enhanced by weighted cooperative representation to obtain the sparse representation coefficient under the dictionary. Then, the test samples were classified and recognized by using the enhancement coefficient and the label matrix of the training samples, and the Gabor feature and weighted cooperative representation classification method were obtained. Experiments on Yale face database, Extended Yale B and AR face database and FERET face database show that the new algorithm has better recognition rate and shorter computation time.