Well defined video object extraction suitable for scalable wavelet based object coding

In this paper we present a semi-automatic multi resolution video objects extraction and tracking algorithm well suited to scalable wavelet based object coding. Objects of interest are determined in the first frame through initial user intervention followed by a spatial segmentation algorithm. The specified objects are afterwards tracked in the subsequent frames. The tracking algorithm includes multiresolution Markov random field (MMRF) based spatial segmentation with emphasis on border smoothness in different resolutions and multi resolution backward partition projection. An intensity change detector indicates newly appeared objects/regions. The proposed method produces well defined and visually pleasing objects as well as allowing for larger motion tracking, better noise tolerance and less computational complexity.