关键词: 稀疏表达, 半耦合, 协同表达分类, 极低分辨率, 人脸识别

Abstract: Existing Super Resolution（SR） algorithms are based on the assumption that High Resolution（HR） and Low Resolution（LR） features share the same intrinsic geometry structure（coupled dictionary learning）. However, the degradation process of LR image makes the features between HR and LR data produce a one to many mapping, which reduces the discriminative information of LR image and decreases the recognition rate of the super-resolution reconstruction image. To solve this problem, this paper introduces the semi-couple sparse dictionary learning model and loosens?up the manifold consistency assumption, while learning a sparse dictionary and a mapping function between sparse coefficients, which improves the consistency of the feature between HR and LR data. On this basis, the introduction of Collaborative Representation Classification（CRC） model achieves efficient classification for semi-couple characteristics. Experimental results show that compared to traditional sparse expression classification algorithm, algorithm in this paper not only improves the recognition rate, and also greatly reduces the time overhead, which verifies the effectiveness of the semi-couple sparse dictionary learning in face recognition.

Key words: sparse representation, semi-couple, Collaborative Representation Classification（CRC）, extremely low resolution, face recognition