Study And Application Of Interactive Image Segmentation

Image Segmentation is a basic technique of image processing, and it is always one of the most difficult techniques in image processing and computer vision. In recent years, the researchers pay more and more attention to the interactive segmentation. On the base of arranging, summarizing and concluding interactive segmentation algorithm, some key theories and approaches in interactive segmentation and its applications are researched in this dissertation. In this dissertation, we attempt to have an in-depth investigation on the Live Wire algorithm, random walk and toboggan method. The main work is as follows:Considering the speed disadvantage of the traditional Live Wire algorithm, a novel Live Wire algorithm based on Pusle Coupling Neural Network (PCNN) is proposed. PCNN is used to obtain the shortest path between the two points by user. The speed of operation is increased by proposed method, especially when dealing with a larger image, the rate reflected a clear advantage.On the study of detection the footprint heel impression, this paper presents an improved Live Wire algorithm by considering the feature of weak edge for heel impression. Ant colony optimization (ACO) algorithm is used to find the shortest path. A new cost function is defined by Sobel operator. The least square method was used for ellipse fitting to obtain the parameters. The improved methods can effectively extract the information of the footprint heel impression.In order to improve the efficiency and speed of random walk algorithm, a new random walk method based on toboggan is presented. Firstly, a graph is created which decomposes the image in scale and space using the concept of toboggan. In this way, we consider each of the regions as the nodes, the weights between graph-nodes is estimated by using the law of universal gravity. Then the label for object and background is drawn by user. Finally, this paper uses the theory of random walk algorithm to segment the image.According to the multi-vehicle problem in Intelligence Transportation System, a vehicle detection algorithm is presented based on the combination of edge feature and random walk techniques. We used background subtraction and edge detection to obtain the moving area, then used morphological operations for vehicle skeleton extraction and get the seeds for random walk. Finally the accurate boundary of moving vehicles is detection by random walk.Considering the over-segmentaion of the toboggan algorithm, two improve toboggan algorithms are proposed in this thesis. Firstly, a new toboggan is presented based on multi-scale morphological gradient. The gradient image is computed by using the multi-scale morphological gradient operation. It was obtained through the difference size of the structure elements from images gradient features. In order to reduce the over-segmentation of toboggan, the approach of region combination is used after toboggan segmentation. Secondly, combining the feedback pulse coupling neural network (FPCNN) with toboggan segmentation algorithm, this thesis presents a new MRI image segmentation feature extraction methods.