Customized RBF kernel graph-cut for weak boundary image segmentation

In recent years, graph-cut algorithms have been considered in image segmentation due to their quality and computational load. Among these algorithms, radius-based function kernel graph-cut approaches with data mapping to new space have yielded good results in image segmentation. However, these approaches for segmenting the weak boundary images, such as cellular images, cannot well distinguish regions with narrow and close boundaries. A customized kernel in a kernel graph-cut algorithm is presented in this paper. In the proposed kernel, the first four terms from the MacLaurin series for sine and cosine functions based on image energy are fused together. Evaluation of experimental results confirms the quality of the proposed algorithm in segmenting the artificial and real image sets with weak boundaries.