A topology independent active contour tracking
Dongxiang Xu, Jenq–Neng Hwang · 2003
In previous years, the active contour (snake) has become one of the most powerful segmentation algorithms in image processing and computer vision. However, most algorithms based on this model have difficulties in automatic initialization and are hard to handle the problems with topology changes or multiple-objects tracking. We propose a model algorithm, quad-tree highest confidence first (QHCF), for image segmentation first. Based on it, a new framework, called Markov random field (MRF) based snake, is then put forward to provide a general purpose image segmentation solution. Since this method combines the most attractive features of MRF and active contour model, it provides more accurate segmentation results. Finally, we further extend this framework to multiple-object scenario and propose a topology independent segmentation algorithm. Experimental results are provided to demonstrate its encouraging performance.