Research On Geometric Active Contour Model And Its Application In Medical Image Segmentation

Image segmentation is a crucial step and important part in image processing, It is abasic and complicant program in the fields of computer vision.It has lots of significantapplications in military, meteorology, medicine and many other fields. The goal ofimage segmentation is to divide a given image into different regions,which theinformation in the same region are similar and better understood,thus other high-leveltasks such as object detection, recognition, and tracking can be further performed,andthe accuracy of segmentation result will impact on the next stage directly.Geometric active contour model based on calculus of variations can deal withtopological change automatically in procedure of curve evolution. It is not sensitive tothe initial curve and easy to implement.It has broad research prospects both in thetheory and application.Geometric active contour model applied in medical imagesegmentation will be concerned in this paper.Data term and regular term in Region-based model,which is one of geometricactive contour model,affect the result differently.Data term will affect the result mainlywhile regular term will determine the length,smoothness,satisfactory of contour.Thesegmentation of the images which have low contrast edges and intensityinhomogeneities is not very accurate but efficient when constants applied in data term,in contrast,substituting functions for constants lead to accurate but not efficient.Becauseof the above,in order to segment intensity inhomogeneity image,constants revised bybias fields fit the parts of image,also a kernel function will be introduced in energyfunctional to balance local and global information of a image.Split-Bregman iterationwill be applied in the model only if regular term equals to length term.For the sake ofglobal optimization,a global convex segmentation model will be conduct beforeSplit-Bregman algorithm applying in.The experiment in medical images demonstrate that:the new data term performswell in intensity inhomogeneity image segmentation,Split-Bregman compute veryfast,but regular term which only contains length term result in satisfactory of contour.