Significance-Preserving-Guided Content-Aware Image Retargeting

With the rapid development of multimedia and network technologies, sharing image contents through heterogeneous devices of different capabilities has been popular. A variety of displays provide different display capabilities ranging from high-resolution computer/TV monitors to low-resolution mobile devices, where images are usually required to be changed in size or aspect ratio to adapt to different screens. Based on the fact that straightforward image resiz-ing operators (e.g., uniform scaling) cannot usually produce satisfactory results, content-aware image retargeting, which aims to arbitrarily change image size while preserving visually prominent features, has been a popular research topic. In this paper, we present a robust and computationally-efficient content-aware image retargeting framework based on seam carving subject to gradient energy and saliency-preserving constraint. In the proposed method, the significance map derived from adaptively integrating both the gradient and saliency maps of an image is used to accurately identify the most important area(s) to be pre-served while retargeting this image. The proposed significance map can well compensate the drawbacks induced by only either gradient-based or saliency-based map is used. As a result, an image can be flexibly adapted to arbitrary sizes. Experimental results demonstrate the efficacy of the proposed algorithm.