Parameter Estimation and Topology Priors in Random Fields
Markov Random Fields (MRFs) are elegant mathematical models for solving many low level vision tasks. One of the most successful use cases of MRFs has been in image segmentation, where the foreground object is isolated from background. This basic framework can be easily extended to videos, where every segmented frame is used as initialization for the next frame. Let us call this MRF based visual tracking. However, there is a critical piece of information that basic MRF based visual trackers ignore, the topology characteristics of objects being tracked. When tracking a human, one knows that a human cannot suddenly split into multiple parts, or suddenly expand into twice the original size. Similarly when tracking a cellular structure, it may be plausible that a cell splits into two cells, or fuses with another to suddenly increase its size. We call this a priori knowledge about the target being tracked as topology priors, and provide a mathematical framework through which MRF trackers are made 'topology aware'
Related Publications
Robust Segmentation based Tracing using an Adaptive Wrapper for Inducing Priors
Vignesh Jagadeesh, James Anderson, Bryan Jones, Robert Marc, Steven Fisher, B.S. Manjunath
IEEE Transactions on Image Processing 2013
Vignesh Jagadeesh, James Anderson, Bryan Jones, Robert Marc, Steven Fisher, B.S. Manjunath
IEEE Transactions on Image Processing 2013