A fast segmentation method based on constraint optimization and its applications: Intensity inhomogeneity and texture segmentation

We propose a new constraint optimization energy and an iteration scheme for image segmentation which is connected to edge-weighted centroidal Voronoi tessellation (EWCVT). We show that the characteristic functions of the edge-weighted Voronoi regions are the minimizers (may not unique) of the proposed energy at each iteration. We propose a narrow banding algorithm to accelerate the implementation, which makes the proposed method very fast. We generalize the CVT segmentation to hand intensity inhomogeneous and texture segmentation by incorporating the global and local image information into the energy functional. Compared with other approaches such as level set method, the experimental results in this paper have shown that our approach greatly improves the calculation efficiency without losing segmentation accuracy.

Research highlights► Proposed a new constrained optimization energy for image segmentation. ► Gave some elegant extension of CVT-based methods to handle intensity inhomogeneous images and texture segmentation. ► Proposed a fast narrow banding algorithm, and mostly outperform its counterparts.