AutoGrowCut — Automatic image segmentation by salient labeling
Jinliang Wu, Gang Wang, Jinyong Chen, Conghuan Tu, Qian Zhang, Qimin Cheng · 2017
Interactive segmentation algorithm has widespread use in image segmentation literature due to its good segmentation results. However, enormous amount of manual labor, namely interactive object and background labeling, is indispensable for a large image database. In this paper, we present an automatic image segmentation method - AutoGrowCut, which is an improved version of GrowCut and uses our visual attention model to obtain the salient labels. Concretely, we propose a new saliency detection model to do saliency map calculation, and adaptive thresholding is conducted on the saliency map to obtain the mask map, which is exploited as initial seed pixels. Then the seed pixels and saliency map are used as the inputs of the algorithm. Finally, the image segmentation process is executed automatically. Experiments with 500 images from MSRA Salient Object Database show that the improved version AutoGrowCut is not only labor-saving, but also has good performance compared to state-of-the-art algorithms.