Image Resizing Based on Self-adapting Significant Map

Abstract

The image resizing methods are roughly categoriz-ed into two classes: simple scaling method and content aware resizing. The former is trivial, while the later is able to preserve visually prominent features in the resized image. The measure approach of visual saliency plays an important role in the content aware resizing method. This paper introduces a new saliency measure, self-adapting significant map, which characterizes the prominent features using gradient, saliency map and resizing ratio. Owing to taking resizing ratio into consideration, our computation approach automatically smoothes the significant map when the resizing is highly non-homogenous. Based on this computation approach, we propose a new content-based resizing method, which can reasonably preserve the important and less-important regions according to the resizing ratio. Experimental results show that our method outperforms the similar resizing method when the resizing ratio is bigger.