Semi - Automatic Video Object Segmentation And Tracking Based On Snake

Building on the rapid development of allied subjects and new developing subjects of the video code, new generation date compression techniques are becoming maturity and reality, the code thought changes from pixel-based to content-based. MPEG-4 has adopted this video code technology of new generation, it expands encoding object from image frame to the meaningful video object for the first time on the development history of the video code, thus realized transforming from the traditional code based on pixel into modern code based on object and content. The primary task of realizing content-based interaction in MPEG-4 is segmenting video / image into different object or extraction of the video object from the background, it is just the video object segmentation and tracking. But because the video object segmentation is a difficult problem with challenge at present, MPEG organization has not made how to cut the existing digital video sequences into video object, but regard it as the open part of the standard and wait for further investigation.Video object segmentation can be divided into automatic segmentation and semiautomatic segmentation according to the user-assisted degree. As to automatic segmentation, these techniques often do not obtain desirable segmentation results because the mathematic model required for the extraction of a video object is difficult to define. So at the present stage, semiautomatic methods are more feasible than the automatic methods.This thesis is focused on semiautomatic video object segmentation and tracking. Based on the in-depth research of active contour models which have been extensively studied and applied as image segmentation methods during the past decades, a framework for semiautomatic segmentation and tracking using Snake is presented.The main research works of this thesis are summarized as follows:1. A new snake model——VSnakes is studied and two improvements of the definition ofcontour energy are proposed. One of them spaces the points more evenly on the contour, another smooth the candidate contour, which not only keeps its shape similar to the current contour, but also make contour more smooth, thus reduce the shape distortion of contour.2. A method of contour prediction is presented which is based on the path coherence and limited contour expand, it can be divided into two courses: global prediction and local modification. The method estimate the position of video object roughly in the next frame using the path coherence firstly, and next step, it is modified as the predict contour by using limited contour expand. The advantage of this method lies in its simple to realize, and it can be used for rigid object and nonrigid object. Combined withsegmentation using Snake which its shrinking effect has been strengthened, the video object segmentation and object tracking can be realized.3. We studied the Open Source Computer Vision Library (OpenCV) which is extensively applied in the image processing and computer vision field abroad at present, in a situation that there are lack of relevant documents yet at home, through exchanging with Dr.Robert Laganier of university of Ottawa and participating in the discussion of OpenCv at Yahoo groups, we use it to implement a large amount of popular algorithms image processing and computer vision, and combine it with DirectShow technology we develop a software about semiautomatic video object segmentation and tracking using Snake.