Image segmentation based on weighting boundary information via graph cut

• Weighting patch is used to describe the boundary information of the image.
• CEMGM algorithm is used to model the region information of the image.
• Weighting parameters are used to combine the region and boundary information.

The graph cut model has been widely used in image segmentation, in which both the region and boundary information play important roles for accurate segmentation. However, how to effectively model and combine these two information is still a challenge. In this paper, we improve the conventional graph cut methods by combining the region and boundary information with an effective and straightforward way. When modeling the region information, the component-wise expectation–maximization for Gaussian mixtures algorithm is used to learn the parameters of the prior knowledge. When modeling the boundary information, the weighting patch is used to represent the similarities of the neighboring pixels. Then the region and boundary information are combined by a weighting parameter, where the weight is small for boundary pixels and is large for non-boundary pixels. Finally, experiments on various images from the Berkeley and MSRC data sets were conducted to demonstrate the effectiveness of the proposed method.