Research And Application Of Image Segmentation Based On Bias Field Variation Level Set Model

In today's society, the digital image have become the essential tool for people to get information and take advantage of them. The image segmentation is one of the most important method in the digital image processing. The meaning of image segmentation is that it is the process of separating the image into several regions according to some similarity criterion. It is the basic of the higher computer vision and digital image processing methods. Therefore, it is very important and necessary to research the image segmentation methods.Up to now, there are so many mature image segmentation methods proposed by the researchers. Because of the superior performance and the flexible structural, the active contour model which basis on the level set method has become the hot spot. The texture widely exist in the digital image. It raises a lot of difficulty of image segmenting because of its special structure. There are three kinds of texture images in our life which includes the nature texture image, the artificial texture image and the hybrid texture images. Basis on the active contour model, we do some research of segmenting the texture image in this paper. The main research content in this paper is partitioned into the following three points:The paper introduce the active contour model and explain the details of the mathematical principles of the level set method. As an example of the region-based level set image segmentation method, the paper introduce the C-V model and point out its limitation in segmenting the intensity inhomogeneity image. Then the variational level set image segmentation based on the bias field was introduced. It could segment the intensity inhomogeneity image well and do well in the medical domain. But it can't get good results when there are some textures in the image. The texture could be restrained by the intrinsic texture descriptor based on texture geometric structure and it would enhance the contrast between different texture regions and smooth the image in the same texture regions. Combining the intrinsic texture descriptor and the bias field variational model, the bias field variational level set image segmentation incorporating restraining the texture was proposed in this paper. The experimental result indicate that the proposed model could segment the natural texture image better compared with the bias field variational model.(1) The bias field variational level set image segmentation incorporating restraining the texture could raise the accuracy of segmenting the natural gray level texture image because of the advantages of the intrinsic texture descriptor. However, when segmenting the artificial texture image or the target and background in the image are so closed, the model couldn't do it well. The structure extraction method via relative total variation make use of a new inherent variation and relative total variation measures could remove the texture elements of the image. By combining it with the bias field variational model, the bias field variational level set image segmentation incorporating image structure extractive model was proposed. The experimental result indicate that the proposed model could segment the artificial texture image better compared with the bias field variational model incorporating restraining the texture.(2) The bias field variational model make use of the weighted K-means clustering method to estimate the bias field of the local neighborhood of each point in the image. It could segment the object region at the same time. Because of its good performance in segmenting the intensity inhomogeneity image, it has a good application in the medical field. By combining the intrinsic texture descriptor and total variation separately, the image segmentation model could do well in the texture images. There are some intensity inhomogeneity and a lot of textures in the remote sensing image because of the influence of a number of factors. It makes against segmenting the target of the. For reducing the influence of the texture information, we use the image segmentation model based on the bias variational field model to segment the remote sensing image in this paper. The experimental result indicate that the image segmentation model based on the bias variational field could segment the remote sensing image better compared with the C-V model.