Structured max-margin learning for multi-label image annotation

In this paper, a structured max-margin learning scheme is developed to achieve more effective training of a large num-ber of inter-related classifiers for multi-label image annota-tion. First, a visual concept network is constructed to char-acterize the inter-concept visual similarity contexts more precisely and determine the inter-related learning tasks au-tomatically. Second, multiple base kernels are combined to achieve more precise characterization of the diverse visual similarity contexts between the images and address the is-sue of huge intra-concept visual diversity more effectively. Third, a structured max-margin learning algorithm is de-veloped by incorporating the visual concept network, max-margin Markov networks and multi-task learning to address the issue of huge inter-concept visual similarity more effec-tively. Our structured max-margin learning algorithm can leverage the inter-concept visual similarity contexts to learn a large number of inter-related classifiers simultaneously and improve their discrimination power significantly. Our exper-iments have also obtained very positive results