Study On Active Contour Models For Images With Piecewise Constant Intensities

Image processing is not only the basic of computer vision, but also an importantpart of image analysis. Image segmentation is a very important prerequisite work and acrucial step of image processing. The goal of image segmentation is to separate thegiven image into a series of meaningful regions, so that we can extract interestingobjects of images. Currently, the image segmentation method based on partialdifferential equations has become one of the most popular research programs due to itsunique advantages. Existing active contour models can be divided into two categories:region-based models or edge-based models. Compared with the edge-based activecontour models, the region-based active contour models have a lot of advantages, suchas: They are independent of the image gradient information, so they can divide weakedge target better; they are using global regional information, so they are more robustwhen dealing with images with noise. This paper focuses on a region-basedsegmentation model to solve the problems that the existing active contour models aresensitive to the initial outline, slow convergence, not split-phase image, need tore-initialize the level set function and so on.In the image segmentation, the essential of many image segmentation methods is tosmooth images and consider the piecewise constant value as the result of thesegmentation. Or consider the piecewise constant value as the initial image and applythe segmentation models to it again to gain a more effective result. Some commonmodels use an image with piecewise constant intensities to approximate the originalimage and or use it to fit the energy function. Therefore, the segmentation of piecewiseconstant values image is of great importance.This dissertation focuses on the applications of region-based models in piecewiseconstant image segmentation, and discusses a series of classical region-based models,such as C-V model, improved C-V model, LCV model and PSM model. The mainworks are summarized as follows:Due to the fact that C-V model is sensitive to initial contours, combines withimproved C-V model and coefficient of variation, an active contour model based onglobal image information is proposed, with which evolution equation is an ordinarydifferential equation. The experimental results validate that the model is not onlyreserved the advantages of the traditional C-V model, but not sensitive to initial contours. So re-initialization is unnecessary. Otherwise, in contrast with some othermodels, the experimental results show that the new proposed model can segment threephases image correctly.