A Study On Video Segmentation And Composition

Segmentation for videos and images is to divide an image into several parts using some criterions, for the purpose to obtain the interested objects. In this thesis, we refer the definition of segmentation to binary segmentation (foreground and background).Segmentation technology has been widely used in fields of digital entertainment, video coding, vision based surveillance, net meeting, video database and so on. So research on the segmentation technology has both important theoretical significance and practical value.The definition of video composition is the operation of realistic combination of two video, with achieving better lighting coherence and smooth transition. Video composition is significant in the film industry and digital entertainment areas, such as reuse of video fodder in film production and realistic digital experience.Based on the systematic survey of the current methods and theories for segmentation and composition, this thesis focuses on the research and implementation of these techniques.In this thesis,we proposes and implements an improved automatic segmentation algorithm and implements an interactive segmentation system, based on the MAP-MRF and Graph Cut framework, with combination of techniques such as Gaussian background modeling, change detection method and background updating algorithm.At the same time, a fast composition method is proposed based on automatic color adjustment and adaptive histogram specification. The experimental results validate our method.The proposed systems of this thesis are developed based on OpenCV with C++ under Windows XP. The experimental results show that the precision can be well achieved in most cases and the systems can achieve outstanding results. .The main contributions of this thesis are as follows.1. Based on the framework of"background cut", we improve the method and propose an automatic segmentation system with better performence.2. Based on color principle, we propose real-time composition method with automatic color adjustment and adaptive histogram specification3. We have implemented an interactive image segmentation system based on the Graph-cut algorithm.