Paper:
Language: English:

Extracting Objects Using Contour Evolutions in Edge-Based Object Tracking

Hiroyuki Tsuji^, Shinji Tokumasu^, Hiroki Takahashi^, and Masayuki Nakajima^

^*Department of Information and Computer Sciences, Faculty of Information Technology, Kanagawa Institute of Technology, 1030 Shimo-ogino, Atsugi, Kanagawa 243-0292, Japan
^**Department of Computer Science, Graduate School of Information Science and Engineering, Tokyo Institute of Technology, 2-12-1 Ookayama, Meguro, Tokyo 152-8550, Japan

Received: February 22, 2005

Accepted: December 21, 2005

Keywords: video object plane (VOP), object extraction, partial differential equations (PDE), contour evolution, level set methods

Journal ref: Journal of Advanced Computational Intelligence and Intelligent Informatics, Vol.10, No.3 pp. 362-371, 2006

Abstract

We propose edge-based object extraction targeting automatic video object plane (VOP) generation in MPEG-4 content-based video coding. In an edge-based VOP generation framework proposed by Meier, the object is represented as a binary edge image that does not generally form a closed contour and that also contains many extra edges, making extracting the object contour accurately less straightforward in such situations. To solve this problem, we adopt a PDE-based contour evolution approach to evolve initial multiple contours contained inside the object toward its boundary based on evolution equations, and to finally merge them into a single contour that accurately represents the object's shape. Our experimental results using an MPEG standard image sequence show that object contours obtained as we propose appear subjectively more natural in shape compared with those obtained by two conventional methods, especially when the binary object model is not in good condition.

[Notice]
* "Preview" is the first 2 pages of the article. You don't need the registration.
* To read the PDF file you will then need to download and install the Adobe Reader.
Adobe Reader is free and available for download here: