Image Segmentation Based on the Hybrid Bias Field Correction

Abstract

Image segmentation is the foundation for analyzing and understanding high-level images. How to effectively segment an intensity inhomogeneous image into several meaningful regions in terms of human visual perception and ensure that the segmented regions are consistent at different resolutions is still a very challenging task. In order to describe the structure information of the intensity inhomogeneous efficiently, this paper proposes a novel hybrid bias field correction model by decoupling the multiplicative bias field and the additive bias field. These kinds of bias fields are assumed to be smooth, so can employ the Sobolev space W1,2 to feature them and use a constraint to the multiplicative bias field. Since the proposed model is a constrained optimization problem, we use the Lagrangian multiplier method to transform it into an unconstrained optimization problem, and then the alternating direction method can be used to solve it. In addition, we also discuss some mathematical properties of our proposed model and algorithm. Numerical experiments on the natural images and the medical images demonstrate performance improvement over several state-of-the-art models.