A fast level set method for inhomogeneous image segmentation with adaptive scale parameter.

It is difficult to segment images in the presence of intensity inhomogeneity due to the overlap of the intensity ranges between different object regions. To deal with this problem, this paper presents a novel level set method to segment inhomogeneous images. Based on the inhomogeneous image model, an optimal segmentation plane is derived in image domain to provide the optimal partition for every pixel. With the plane, a new region-based pressure force function is proposed and used to define an energy functional in the level set formulation on the whole image region. By minimizing the energy functional, the proposed method can segment the inhomogeneous image and estimate the bias field at the same time. Besides, a new bias field initialization is introduced to improve the robustness to the initial contour. In addition, a novel adaptive scale parameter is designed for the kernel function in order to estimate the bias field accurately. The proposed method is first presented as a two-phase level set formulation and then extended to a multi-phase one. Finally, the experimental results on both synthetic and real images demonstrate the superiority of the proposed method in terms of accuracy, efficiency and robustness.