Motion-based visual object extraction in video

Video content analysis is an important step in video cataloging while the visual objects contribute significantly to comprehension of the content. Many efforts have been made to extract meaningful region in the video. We propose a scheme to extract visual objects based on motion attributes of the object. The scheme uses a wavelet-based optical flow. A 2.5D motion model is proposed to analyze the general motion of the frame. With a,compensation of general motion we segment each moving object using a competitive agglomeration scheme and representing regions as a number of layers. Keywords: video analysis, content-based video retrieval, multimedia content characterization.