Video Object Segmentation Based on Motion Information and Region Filling

• Main Authors: Rong-Yu Jheng, 鄭榮裕
• Other Authors: Cheng-Chin Chiang
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/82839978784168001241

碩士 === 國立東華大學 === 資訊工程學系 === 93 === In recent years, among various emerging multimedia technologies the MPEG-4, an object-based multimedia standard, continuously attracts the public attention from both academic and commercial realms. For multimedia video, encoding the contents in units of video objects generally gains better performance than the conventional block-based encoding does. In this thesis, we propose a novel method to accomplish the task of video object segmentation. The key concept is to analyze the object motion from multiple consecutive frames along the time line and build masks for moving objects. Afterwards, accurate contours of moving objects can be obtained by integrating the built motion masks and the gradient image calculated from the current frame. Finally, the proposed method applies techniques of region filling and contour refining to extract the exact video objects from the current frames. The experimental results show the effectiveness of the proposed video object segmentation method.