Modified localized graph cuts based active contour model for local segmentation with surrounding nearby clutter and intensity inhomogeneity

Graph cuts based active contour models (GCACMs) have been widely used in image segmentation for global minimization and efficient calculation. For local segmentation, the localized GCACMs (LGCACMs) have been proposed by reformulating GCACMs in a narrow band. However, the existing LGCACMs cannot deal with local segmentation with surrounding nearby clutter and intensity inhomogeneity. In this paper, a modified LGCACM (MLGCACM) with a new narrow band energy function is proposed to solve local segmentation in the presence of surrounding nearby clutter and intensity inhomogeneity. Firstly, by removing the region term in the narrow band energy function, the narrow band graph can be constructed without t-links, in which case, local segmentation with surrounding nearby clutter can be solved. Secondly, by strengthening the edge term in the narrow band energy function with a new local region term, the n-links in the narrow band graph can be weighted more suitably for local segmentation with intensity inhomogeneity. Experiments on synthetic and medical images demonstrate the advantages of the proposed MLGCACM over the existing LGCACMs in local segmentation with surrounding nearby clutter and intensity inhomogeneity.