分块稀疏表示的人脸识别研究

Abstract

针对人脸识别中人脸被局部遮挡对识别效果带来的严重影响，提出一种对遮挡具有鲁棒性的分块稀疏表示分类的人脸识别算法。稀疏表示分类算法利用高维数据分布的稀疏性进行建模，能够很好地解决图像高维处理问题，有效地避免维数灾难。通过对该算法进行改进，提出一种分块稀疏表示的人脸识别算法。首先对人脸图像进行分割，独立地对每个子块进行稀疏表示分类，再通过所有子块的分类结果进行联合判别。改进后的算法避免了特征提取过程中所造成的图像信息丢失，也避免了人脸部分信息丢失对整体识别结果的影响。通过在AR和Yale人脸数据库上进行仿真实验，可以得出该改进算法能显著提高遮挡人脸图像的识别率，并且对光照变化也具有一定的鲁棒性。 In order to reduce the sensitivity of the face recognition algorithm to occlusion, a robust occlusion block sparse representation classification face recognition algorithm is proposed. The sparse representation algorithm uses the sparsity of high-dimensional data distribution to perform modeling, which can deal with high-dimensional image and effectively avoid dimension disaster. Block thinking is introduced in this algorithm. First of all, face image is divided into blocks which are independently sparse representation classification, and then a joint determination by all classification sub-blocks. The improved algorithm not only avoids the image feature extraction process information loss caused, but also avoids the loss of face parts information on the overall recognition results. Through simulation experiments on AR and Yale face database, it can be drawn that the improved algorithm can significantly improve the recognition rate of occluded face image, and also have some certain robustness under variable illumination.