This paper investigated the issue of image change detection
based on MRF models. These models characterize the statistical
correlation of intensity levels among neighboring pixels more
accurately than pixel-based models We have developed a new
ICD algorithm based on an MRF model that employs the MAP
criterion. The algorithm involves the search for an optimum for
which the SA algorithm is used. By means of two examples, we
have shown the superior performance of our algorithm. This is
due to the use of contextual information as well as the computation
of the true MAP solution. The effect of uncertainties on
the performance of our algorithm was also investigated. Noise,
misregistration, and modeling errors were considered to be the
sources of uncertainty. Our algorithm was found to be quite robust
to various types of uncertainties. We realize that the independence
assumption of the configurations of changed sites
given the configurations of unchanged sites may not be suitable
for all cases. A more accurate model may result in better performance
and may be considered in the future