This paper presents a semi-automatic algorithm for video object extraction, in which a semantic object of interest is defined in advance in a key frame provided by human. For ordinary video frames, the specified video object is tracked and segmented automatically using Learning Vector Quantization (LVQ). This paper also presents a technique for improving extraction performance of the proposed algorithm and reducing computation time. Experimental evaluation using MPEG standard test video sequences demonstrates that the proposed algorithm is able to extract the video object with low error.
An Improved Technique for LVQ-Based Video Object Extraction