A Study Of Image Segmentation Of Statistical Information Integrated Geodesic Model

Active contour model is a big breakthrough in the field of imaging segmentation,which is able to integrate the image boundary and regional information, express a prioriknowledge, and with feasible algorithm support. Active contour model has been widelyused and rapidly developed in edge detection, medical image segmentation and motiontracking. It has become the most active research topic in the computer vision fieldrecently.This paper introduces several existing and classic active contour models firstly,then analyzes the advantages and disadvantages of each model, finally compares eachmodel’s experimental results. Through analysis it is found that the Chan-Vese modelwhich adopts the global gray information of images is not sensitive to the contourinitialization and noise, but the model cannot segment images with intensityinhomogenity; the local binary fitting model using the local gray information of imagescan segment intensity inhomogeneity images, while the model is sensitive to contourinitialization and has large amount of calculation. To solve these problems this paperputs forward a kind of improvement image segmentation algorithm. Firstly thisalgorithm cans statistics information integrated geodesic model. The global and localgray information is integrated into signed pressure force function; making two kinds ofgray information better fuse and target boundary be captured better. Then while thesigned pressure force function’s simple realizing method can reduce the operation costof the algorithm. Secondly the thesis analyzes the global and local gray informationinfluences on the process of model evolution. The research finds that the localinformation will play absolute role when the target is near to the border, and the globalinformation will take absolute role when the goal is away from the border. Thenaccording to the characteristics the paper constructs membership functions usingimage’s local information. This function can determine the model’s combinationcoefficient of the global and local gray information to improve the applicability of themodel. Finally, the paper makes the experiment on different types of images. Thesegmentation results show that this method has the following advantages: segmentingimage with intensity inhomogenity well; being not sensitive to contour initialization position; simple realizing process resulting in making the iterative times and timesignificantly less to get results; having good effect on the improvement method ofcombinatorial coefficient.