基于带补偿字典的松弛稀疏表示的小样本人脸识别*

在训练集类内变化类型不可控的小样本人脸识别问题中，补偿字典很难发挥足够作用。在基于带补偿字典的稀疏表示的人脸识别方法中，训练集字典和补偿字典对测试图片表示的能力不同，文中讨论因此不同而导致的二者在稀疏性上的不同要求，通过对两类字典采用不同的稀疏性约束，提出基于带补偿字典的松弛稀疏表示的人脸识别方法。实验表明，在训练集图片类内变化类型不可控的小样本人脸识别问题中，文中方法能取得较优效果。

Loose Sparse Representation Based Undersampled Face Recognition with Auxiliary Dictionaries

In the undersampled face recognition problem with uncontrolled intra-class variations, the auxiliary dictionary can not work quite well. The training dictionary and the auxiliary dictionary in the sparse representation face recognition methods have different representation abilities for the query image. Thus, different demands on the sparsity constraints of these dictionaries at representation stage are discussed. In this paper, a loose sparse representation based classification with auxiliary dictionaries (LSRCAD) is proposed by using different constraints on two types of dictionary respectively. The experiments confirm the effectiveness and the robustness of LSRCAD. LSRCAD outperforms the original sparse representation face recognition methods with auxiliary dictionaries for undersampled face recognition problems.