Image Retargeting Based on a New Salient Region Detection Method

Abstract

Image retargetting aims at automatically adjusting the size of a given image to fit the arbitrary displays without introducing severe visual distortions. Salient detection plays a key role in image retargeting to preserve visually important image content. We propose a new salient region detection method and an improved seam carving scheme for image retargeting. Our main contribution includes: we design a convex hull of salient points to detect salient regions, we also propose a weighted image importance map (IIM) to measure importance of pixels. This IIM can highlight salient regions while suppressing the saliency of background. This helps to preserve salient objects well and makes the resized image have better visual effects. In addition, we present a method to select the scale parameter of cartoon-texture decomposition used in preprocessing. Experimental results demonstrate the proposed approach is more effective in image retargetting compared to other state-of-the-art methods.