Interactive image segmentation via multi-cue dynamic integration

Due to the complex scenes involved in the natural images, some interactions with the user are necessary to obtain meaningful segmentation results. Interactive image segmentation is such a processing technique which extracts objects from a complex image background according to the user's demand. There are many intrinsic attributes affecting the result of segmentation. It is hard to produce a good result based on a single clue of the image. This paper presents a novel image segmentation method based on the dynamic multi-cue integration. This method fuses color, texture, spatial and edge information via a conditional random field. Most existing image segmentation algorithms combine multiple cues statically with a constant ratio, it is hard for them to describe the inherent property of different images efficiently. The proposed method takes the full advantage of the user labeled information, and constructs a conditional random field containing different energy terms. These energy terms are fused dynamically in accordance with the image inner property, thus greatly improves the segmentation power of the model. In addition, we use the Jensen-Shannon divergence to extract the border information, which is robust against noise. The experimental results show the proposed method performs steady and works well on various natural images.