Gac Model-based Interactive Image Segmentation Algorithm And Its Application

Image segmentation, which aims to separate the interested object of the image from the other parts so as to serve the upper level of the image processing ,is a very important part of image processing. Because of the limitation of the traditional method, image segmentation is difficult to meet people's requirements. So it's hoped that the introduction of human-computer interaction to solve the ill-posed of traditional method. In recent years, interactive image segmentation has been greatly developed, and has been widely used in practice.This thesis summarizes some interactive image segmentation methods in common use, and compares their performance. We can see the traditional theory and methods of image segmentation are the foundation of the interactive image segmentation. First, it introduces some edge detection methods, and evaluates their performance. Then, it describes the basic theory of the level set method and its application in image segmentation field.Based on GAC model, this thesis advances an improved algorithm for interactive image segmentation. First, it adopts the Total Variation model to pretreatment the image so that the edge of the image is well protected while denoising. Second, it adopts an edge detection algorithm according to the curvature-weighted gradient module to acquire aggregate of the points in boundary. Third, it recommends these points of the bigger curvature as candidate ones to the user. Finally, the user chooses appropriate points according to their judgments or directly chooses some ones in front of candidate points, thus the computer will achieve the object's segmentation on GAC model.We develop a simple graphic user interface for the sake of the operation convenience, which allows the users to select the points from a list. The system not only can make the accuracy and consistency of results guaranteed, but also greatly reduce the excessive demands for the operation of the user experience and skills compared to the original method.Experiments are given to show that the improved algorithm significantly enhances the automatization level of interactive image segmentation, and effectively reduces workload in the interactive course.