作　　者： (赵雯）; (吴小俊）;

机构地区： 江南大学物联网工程学院,江苏无锡214122

出　　处： 《计算机应用研究》 2017年第10期3157-3161,共5页

摘　　要： 针对人脸识别中的图像存在噪声等情况,提出基于鉴别性低秩表示及字典学习的算法。使用鉴别性低秩子空间恢复算法(DLRR)获得类别间尽可能独立且干净的训练样本,然后通过引入基于Fisher准则的字典学习(FDDL)方法得到结构化字典,其子字典对对应的类有较好的表示能力,约束编码系数具有较小类内散列度和较大类间散列度。最后对测试样本稀疏线性表示时正确类别的样本贡献更大。在标准人脸数据库上的实验结果表明该算法有较好的性能。 In consideration of noises appeared in the images used in face recognition, this paper proposed a novel recognition method of discriminative low-rank representation with dictionary learning. It used a discriminative low-rank representation（ DL- RR） method to obtain a clean dictionary whose sub-dictionaries for distinct classes required to be as independent as possible. By introducing a Fisher discrimination dictionary learning（FDDL） method, it could obtain a structured dictionary, and each sub-dictionary in the whole structured dictionary had good representation ability to the training samples from the associated class. The coding coefficients had small within-class scatter but big between-class scatter. Finally, the samples from the correct class did greater contributions to the sparse linear representation of the test sample. The experimental results on the standard face databases show that the proposed method has good performance.

关 键 词： 人脸识别 低秩表示 字典学习 稀疏线性表示