Recent Changes - Search:


edit SideBar


Optimal Probabilistic Relaxation Labeling

I. Poole, Proc. BMVC 1990, BMVA, 1990.

Many people use relaxation labelling based approaches to refine spatial distributions of class labels. This paper is the only one I have seen which explains the importance of doing this in a quantitative statistical framework. In particular the concept of the `honest' probability is used as a fundamental principle in the design of the update scheme. The work illustrates that conventional methods, which assume a fixed update function, cannot be honest and therefore also not valid use of probability theory. In my view this paper sets the standard for all work in data fusion and regional analysis. Unfortunately, shortly after completing this work Ian Poole completed his PhD and left academia, so the work was not submitted to a journal. If you can find it however, it is well worth reading. How often do you read a paper which gives you a new (valid) design principle?

NAT 8/3/2005

Edit - History - Print - Recent Changes - Search
Page last modified on March 05, 2012, at 03:52 PM