Graph Cuts based image segmentation using local color and texture

Abstract

this paper proposes a novel interactive image segmentation algorithm based on the Graph Cuts by employing the color and texture feature of pre-computed over-segmentation region. Firstly, Watershed algorithm based on color information has been used to partition the image into a lot of different regions which will be considered as the nodes of Graph Cuts, instead of image pixels. Then the color and texture features (uniform local binary pattern) were extracted from the aforementioned regions, which are used to redefine the region properties term and boundary properties term. Due to pretty integrate the color feature with texture feature; the proposed method can separate the object from its background with colors similar to its background. Experiments on Berkeley image datasets demonstrate the effectiveness of the proposed method.