A Variational Level Set Approach Based on Local Entropy for Image Segmentation and Bias Field Correction.

Jian Tang,Xiaoliang Jiang

doi:10.1155/2017/9174275

Jian Tang, Xiaoliang Jiang

Open Access

https://doi.org/10.1155/2017/9174275

Copy DOI

Abstract

Image segmentation has always been a considerable challenge in image analysis and understanding due to the intensity inhomogeneity, which is also commonly known as bias field. In this paper, we present a novel region-based approach based on local entropy for segmenting images and estimating the bias field simultaneously. Firstly, a local Gaussian distribution fitting (LGDF) energy function is defined as a weighted energy integral, where the weight is local entropy derived from a grey level distribution of local image. The means of this objective function have a multiplicative factor that estimates the bias field in the transformed domain. Then, the bias field prior is fully used. Therefore, our model can estimate the bias field more accurately. Finally, minimization of this energy function with a level set regularization term, image segmentation, and bias field estimation can be achieved. Experiments on images of various modalities demonstrated the superior performance of the proposed method when compared with other state-of-the-art approaches.

Highlights

Image segmentation has always been a crucial step in image understanding and computer vision
According to the observed signal model of the image with intensity inhomogeneity, the local Gaussian distribution fitting (LGDF) energy function that includes the local entropy is defined for driving the evolution contour of the level set toward the desired boundary
Li’s model [8], Zhang’s model [9], Wang’s model [10], RSF model [20], LGDF model [21], and the method of this paper are applied on a variety of synthetic images and medical images

Summary

Introduction

Image segmentation has always been a crucial step in image understanding and computer vision. In [10], Wang and Pan proposed a new image-guided regularization to restrict the level set function In this method, tissue segmentation and bias field estimation are unified into a single Bayesian inference framework and are simultaneously achieved by minimizing the objective energy functional. According to the observed signal model of the image with intensity inhomogeneity, the LGDF energy function that includes the local entropy is defined for driving the evolution contour of the level set toward the desired boundary. The means of this objective function have a multiplicative factor that estimates the bias field in the transformed domain. (2) By incorporating the local entropy information, our model can estimate the bias field more accurately

Background

The Proposed Scheme

Experimental Results

Conclusion