基于局部通用的分块协同表示的单样本人脸识别

Local Generic Based Patches Collaborative Representation for Single Sample Face Recognition

针对单样本人脸图像存在的光照、表情、以及遮挡变化,提出一种局部通用的分块协同表示单样本人脸识别算法。算法首先根据训练样本构建通用的训练字典,建立自然变化下的人脸和部分为因光照、表情遮挡等变化的通用人脸集。其次将测试样本、训练样本进行分块,构造字典块矩阵。通过正则化协同表示对样本块进行稀疏表示和分类。在Extended Yale B、AR公开的人脸数据上仿真实验,这种基于通用表示的分块协同表示单样本人脸识别算法具有很好的鲁棒性,在训练样本较少以等复杂环境变化下,不论是识别效果还是计算复杂度都取得很好的效果。

This paper proposes a local generic based patches collaborative representation for single sample face recognition with illuminations, expressions and disguises. First of all, according to the training sample build joint dictionary, natural face and changes such as illumination, expression and disguises, secondly partition test samples and training samples into patches, structure patch dictionary. The classification is performed based on the residual through Collaborative representation.The experiments on Extended Yale B, AR face data show that local generic representation based patches collaborative representation for face recognition has good robustness. It shows outperforms than many state-of-the art methods on recognition effect and computational complexity under complex environment changes.