New semi-supervised classification using a multi-modal feature joint L21-norm based sparse representation

In this paper, a novel semi-supervised classification method, namely sparse semi-supervised classification algorithm (SSSC) is proposed. To improve the reliability of SSSC, this study extends SSSC to multi-modal features joint L21−norm based sparse representation. In the SSSC framework, the labeled patterns are sparsely represented by the abundance of unlabeled patterns, and then the scores of the unlabeled patterns are computed corresponding to the object class based on the relational degree vector. A quality measure is also presented to divide the unlabeled patterns into reliable and unreliable relabeled patterns. The reliable relabeled patterns are selected to be added into the labeled data for learning the labels of the unreliable relabeled data recurrently. Experimental results clearly demonstrate that the proposed method outperforms the state-of-the-art classification methods. •We develop a new semi-supervised classification algorithm based on a multi-modal feature joint L21-norm sparse representation.•In the proposed optimization, the labeled patterns are sparsely represented by the abundant of unlabeled patterns, then the scores of the unlabeled patterns are computed corresponding to the object class based on the relational degree vector.•A quality measure is also presented to divide the unlabeled patterns into reliable and unreliable labeled data set. The reliable labeled patterns are selected to add into the labeled data to learn the labels of the unreliable labeled data recurrently.