Image patch partition Image-width-fixed retargeting by interactively assigning different patchwise scaling factors Our retargeting result MULTIOP SC WARP SCL SV Figure 1: Patchwise scaling for image retargeting. The first row: examples of retargeting the BOAT image in RetargetMe benchmark to half size using six methods. The second row: Retargeting the TAJMAHAL image with the fixed-image-width constraint. Abstract Content-aware image retargeting methods have recently received increasing

more »

... s. In this paper we improve a patchwise scaling method for image retargeting at an object level. The improvements include a simple yet effective patch partitioning scheme and an optimal scaling factor assignment algorithm. The improved patchwise scaling method first takes the overall image structure into consideration by partitioning the image into rectangle patches of adaptive sizes, which are comparable to the sizes of salient objects in the image. This partitioning is based on a visual saliency map and accordingly the partitioned patches are labeled important and non-important. Then an optimal patchwise scaling method is applied that scales the important patches as uniform as possible and stretches/squeezes the non-important patches to fit the target size. To find an optimal set of scaling factors, a patch-based image similarity measure is proposed to guide the optimization process. Experimental results show that the improved patchwise scaling method has a good performance in image types of lines/edges, foreground objects and geometric structures.