SVM based multi-label learning with missing labels for image annotation

Recently, multi-label learning has received much attention in the applications of image annotation and classification. However, most existing multi-label learning methods do not consider the consistency of labels, which is important in image annotation, and assume that the complete label assignment for each training image is available. In this paper, we focus on the issue of multi-label learning with missing labels, where only partial labels are available, and propose a new approach, namely SVMMN for image annotation. SVMMN integrates both example smoothness and class smoothness into the criterion function. It not only guarantees the large margin but also minimizes the number of samples that live in the large margin area. To solve SVMMN, we present an effective and efficient approximated iterative algorithm, which has good convergence. Extensive experiments on three widely used benchmark databases in image annotations illustrate that our proposed method achieves better performance than some state-of-the-art multi-label learning methods.