A new tracking mechanism for semi-automatic video object segmentation

Abstract

This paper proposes a new tracking mechanism for semi-automatic video object segmentation. An interactive video object segmentation tool is presented for the user to easily define the desired video objects in the first frame, and then the video objects are automatically segmented using the proposed tracking mechanism of bi-directional projection. Forward projection is first exploited to locate the current video object with rough boundary information. Watershed segmentation is then applied to the simplified gradient image of the current frame to obtain a reasonable partition. An improved backward projection, which incorporates pixel classification with region classification, is finally performed on some segmented regions in a rather small search range, and the tracking performance is enhanced in respect of both reliability and efficiency. Experimental results for various types of the MPEG-4 test sequences demonstrate an efficient and faithful segmentation performance of the proposed approach. © Springer-Verlag Berlin Heidelberg 2004.