Selecting suitable image retargeting methods with multi-instance multi-label learning

Abstract

Althogh the diversity of mobile devices brings in image retargeting technique to effectively display images on various screens, no existing image retargeting method can handle all images well. In this paper, we propose a novel approach to select suitable image retargeting methods solely based on original image characteristic, which can obtain acceptable selection accuracy with low computation cost. First, the original image is manually annotated with several simple features. Then, suitable methods are automatically selected from candidate image retargeting methods using multi-instance multi-label learning. Finally, target images are generated by the selected methods. Experiments demonstrate the effectiveness of the proposed approach. © Springer International Publishing 2013.