Technique Study Of Image Segmentation Based On Active Contours

Image segmentation is an important technique in Digital Image Processing. Themain purpose of image analysis is to analyze and process the target of interest of theimage. With the development of scientific and technological, the technology of imagesegmentation is more and more important in modern life. Compared with thetraditional image segmentation algorithm, the algorithms based on level set evolutionhave many advantages: it has completed segmentation boundaries and the imageadditional information can be integrated into this kind of algorithms. So activecontours have become a hotspot in image study, and have gotten much attention ofmany scholars. This paper introduced several classical active contour models, throughanalyzing the intrinsic traits of these models, some improve algorithms were proposedto solve their existing problems.Level Set Evolution Without Re-initialization (LSEWR) is a classic activecontours, but it is too dependent on the gradient of the image, resulting in a undesiredresult for noise image segmentation, so this paper proposed a novel active contourmodel for image segmentation. By using the minimum measurement of multi-scaleneighborhood inhomogeneity to construct the edge stopping function instead of thegradient, this model can have a better adaptability. By making lots of experiment, thismethod has a good anti-noise performance and high evolution speed by experimentalresults.In addition, this paper also introduced several classic models based on regioninformation, such as Mumford-Shah model, CV model, LBF model. Throughanalyzing the intrinsic traits of these models, this paper proposed a improved model,which combines the global information, local information and boundary informationof the image. In this model, the local information of the image is used to segment thein-homogeneity images,the global information is used to avoid this model fall intolocal minimum, the boundary information is used to accelerate the evolution speed.Many experimental results had proved effectiveness of this model.