Region Merging Based Segmentation with Cellular Automaton
Yong Gao, Sheng Zheng Wang, Jie Yang · Applied Mechanics and Materials · 2013
Since fully automatic image segmentation on natural images is usually hard to provide guaranteed results, interactive scheme with a few simple user inputs becomes a good alternative. This paper presents a novel interactive method based on regional attacking and merging mechanism within a cellular automaton (CA) framework. With an attacking rule based on regions maximal similarity, the adjacent homogeneous regions that are initialized by pre-segmentation are automatically merged and labeled, the users only need to indicate the object and background regions with rough markers. The whole process neednt set any similarity threshold in advance and the desired contours are effectively extracted by labeling all the non-marker regions as either background or object. Extensive experiments are performed and the results show that the proposed scheme can reliably extract the object contours from the complex background.