Multi-layer supervised dictionary learning for visual classification

Abstract

For the task of visual categorization, the learning model is expected to be endowed with discriminative visual feature representation and flexibilities in processing multi-layer categories structure. Many existing approaches are designed based on a flat category structure, or rely on a restricted category structure, hence may not be appreciated for dealing with complex category structure and large numbers of categories. In this paper, we propose a novel dictionary learning method by taking advantage of the hierarchical category structure. A shared discriminative dictionary and a discriminative classification model are learnt for visual categorization. An optimization framework for learning all the components of the proposed model is presented. In the process of optimization, the hierarchical semantic structure among categories is preserved in the dictionary. Experiments on Caltech256 and ImageNet object data subset demonstrate that our approach achieves promising performance on data with large numbers of classes compared with some state-of-the-art methods.