The Weighted Least-Squares (WLS) estimation of binary Markov Random Field parameters is presented. The properties which makes it very useful in image analysis are given. The combination of more robust estimation, checks on estimation accuracy, model order and parameter confidence limits, and speedy execution are essential in applications where much of the estimation may have to be unsupervised, or where very large data sets have to be processed. In these experiments, the performance of ML and WLS estimators over a large volume of the parameter space rather than a few sample vectors was considered, thus giving a better examination of the effects of numerical stability on likely normal performance.
|IEE Colloquium (Digest)
|Published - 1995
|IEE Electronics Division Colloquium on Multiresolution Modelling and Analysis in Image Processing and Computer Vision - London, UK
Duration: 21 Apr 1994 → 21 Apr 1994