Topology structure based saliency region detection for cartoon images
Rao Zeng, Juncong Lin, Xing Hua Gao, Li-sheng Xiang, Minghong Liao · 2014
Saliency detection is a key component of content-aware image processing, such as image retargeting. The speciality of cartoon images make the existing algorithms hard to get excellent results. In this paper, we propose a new method to detect saliency region of cartoon images, based on the topology structure of superpixels. Firstly, we extract the feature lines and superpixels of the cartoon image and establish the topology structure of superpixels to indicate the relationship between each of them. Secondly, we extract the superpixels adjacent to the extracted feature lines and distinguish the background superpixels from the foreground superpixels through heuristic search on the topology structure. Lastly, the global saliency is computed by calculating the sum of color distance between the background superpixels, while local saliency is computed by the a novel Saliency flood scheme. The experimental results demonstrate that our algorithm outperforms recent state-of-the-art saliency detection methods on cartoon images, yielding higher precision and better recall rates.