Study On Image Segmentation Based On Snake Model

In the past recent twenty years, Snake model, as an important research filed in computer vision, attracts the attention of more and more researchers. It is a top-down processing with prior knowledge, which is different from the classical computational vision proposed by D. Marr. It provides a theoretically uniform framework to a series of problems, such as contour extraction, stereo matching and object tracking. It also has been successfully applied to image segmentation, medical image processing, human-computer interaction and many other research and practical fields. However, there are still some problems and limitations when it is used in practice, thus further study on it to propose an improved one is a necessity.This paper describes the traditional Snake model, introduces its fundamental theory and its numerical solution, presents some improved Snake models, systematically and profoundly analyzes and summarizes them as well as their advantages and disadvantages. And then, aiming at the problems of iris location in biometric recognition, the paper presents two improved snake models to respectively locate the inside and outside iris edge. When locating the inside edge, the gradient image of iris without being filtered was first directly used, then the normalized vector normal to each snake node pointing to the inner iris edge was added and this improved snake can locate the iris inside edge under these two factors; when locating the outside edge, based on the traditional snake model, the normalized vector at every snake node pointing to the centroid of the contour was added to help. First the centroid of snake was calculated, and then normalized vector between each snake node and the centroid pointing to the centroid was computed, with the help of these two factors, the snake can locate the iris outside edge successfully.Large amount of experiments are carried out and they show that the proposed snake is robust and comparing to the classical iris location algorithms such as the algorithm of Daugman and Hough Transform combined with edge detection operator, the proposed models has a more accurate result.At last, the three main parameters of Snake model are studied through experiments to see the effects to the Snake when they are changing. The paper discusses how to value them and provides resources and references for their further design.