Kyoung-Rok Lee and Truong Q. Nguyen

Video Processing Lab

University of California, San Diego


Object extraction and tracking are critical operations for many content-based video applications. This paper proposes an efficient and accurate object extraction and tracking method without any user interaction or extra information. Our approach is based upon extracting objects in the form of components from the scene using a spectral matting method, and tracking these components using component correspondence.



1. Object mattes from Aya sequence (frame 16)


2. Video matting process

Frame 15
Frame 16
Frame 17

Original image

Motion map

Object matte


3. Composite images

Composite (Nana sequence)
Composite (Aya sequence)