A Hierarchical Scale-and-Stretch Approach for Image Retargeting

Abstract

Automated image retargeting techniques are becoming important with the proliferation of different display units, such as cell phones, notebooks, televisions etc. Scale-and-stretch techniques have been successfully used for resizing images into different aspect ratios, while also preserving the most prominent visual features. The main idea of scale-and-stretch techniques is to utilize a single grid with predefined resolution, and map onto it the single significance map. The problem of image resizing is subsequently formulated as an optimization problem, which determines the optimal deformation for every local region of the image based on their underling significance. The use, however, of a single grid with specific distribution is not robust with respect to the size of the significant regions that exist in the image. In this respect, this paper extents the scale-and-stretch techniques with the use of a hierarchical grid that incorporates the significance maps of different grid resolutions in a single grid. The proposed hierarchical approach surpasses current methods and manages to efficiently identify significant regions and achieve better results.