An Efficient Block-Based Video Segmentation with Automatic Renewal Scheme

• Main Authors: Pei-Ling Huang, 黃佩玲
• Other Authors: Kuo-Liang Chung
• Format: Others
• Language: en_US
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/98008102461060358172

碩士 === 國立臺灣科技大學 === 資訊工程系 === 92 === Video segmentation aims to segment moving objects in video sequences. After segmenting objects in each image frame, these segmented objects have many applications, such as object manipulation, scene composition, and MPEG-4 standard compression. In this thesis, we proposed an efficient block-based algorithm for video segmentation. Specially, a region merging strategy is employed to alleviate the over-segmentation problem. Based on an edge-based accumulation error criterion, an automatic renewal scheme is presented to restart the spatial segmentation for the next image frame as the I frame to interrupt the error propagation problem. Under some real video sequences, experimental results demonstrate that our proposed block-based video segmentation algorithm is encouraging when compared to the previous two video segmentation algorithms---the one by using the watershed technique to segment independently each image frame in the video sequence and the block-based predictive watershed algorithm proposed by Chien, Huang, and Chen.