Abstract

Content-based representation of video sequences for applications such as MPEG-4 and MPEG-7 coding is an area of growing interest in video processing. One of the key steps to content-based representation is segmenting the video into a meaningful set of objects. Existing methods often accomplish this through the use of color, motion, or edge detection. Other approaches combine several features in an effort to improve on single-feature approaches. Recent work proposes the use of object trajectories to improve the segmentation of objects that have been tracked throughout a video clip. This thesis proposes an unsupervised video object segmentation method that introduces a number of improvements to existing work in the area. The initial segmentation utilizes object color and motion variance to more accurately classify image pixels to their best fit region. Histogram-based merging is then employed to reduce over-segmentation of the first frame. During object tracking, segmentation quality measures based on object color and motion contrast are taken. These measures are then used to enhance video objects through selective pixel re-classification. After object enhancement, cumulative histogram-based merging, occlusion handling, and island detection are used to help group regions into meaningful objects. Objective and subjective tests were performed on a set of standard video test sequences which demonstrate improved accuracy and greater success in identifying the real objects in a video clip compared to two reference methods. Greater success and improved accuracy in identifying video objects is first demonstrated by subjectively examining selected frames from the test sequences. After this, objective results are obtained through the use of a set of measures that aim at evaluating the accuracy of object boundaries and temporal stability through the use of color, motion and histograms