Multi-layer graph constraints for interactive image segmentation via game theory

The combination of pixels and superpixels has been widely used for image segmentation, where the pixels and superpixels are segmented together. These combination methods can obtain more robust results by using more informative superpixel features. However, since the superpixel may not accurately capture the details for the small and slender regions, the results of these combination methods are often label inconsistent with the objects. Furthermore, these methods also fall into expensive time cost due to introducing more interactions between pixels and superpixels. To overcome the above problems, in this paper, we propose an interactive image segmentation method based on multi-layer graph constraints. The relationships between pixels/superpixels and labels are introduced into the conventional combination framework to further improve the segmentation accuracy. The segmentation model is constructed based on the estimation of probabilities of pixels and superpixels by a nonparametric learning framework. Then the probabilities of pixels and superpixels are updated iteratively by utilizing the game theory based optimization strategy. Experiments on challenging data sets demonstrate that the proposed method can obtain better segmentation results than the state-of-the-art methods. (C) 2016 Elsevier Ltd. All rights reserved.