C.T. Li and R.G. Wilson, Multiresolution Image Segmentation Using Markov Random Field (September 1, 1996).
In this work we propose a statistical multiresolution approach by modelling spatial gray level difference using Markov Random Field (MRF). Firstly, a multiresolution gray level pyramid is built by applying a unimodal Gaussian-like kernel operator to subsample the original image by a factor of 2. With the original image (highest spatial resolution and lowest class resolution) at the bottom of the pyramid, each level is a low-pass/sub-sampled version of the level below. Subsequently, based on Markov Random Field, the measure of grey level variances of pairs of pixels within region and across regions are utilised to estimate the maximum a posteriori classification at the top level of the pyramid (highest class resolution and lowest spatial resolution). Stochastic Relaxation is also adopted iteratively to maximise the likelihood. After the class partitioning is done at the top level the information gathered is propagated down to the next level and the algorithm is applied again to the new level to refine the position of the boundary. Results are presented on various test images to show the effectiveness of the method.
<%@ include file="cited.html" %>C.-T. Li and R.G. Wilson, "Textured Image Segmentation Using Multiresolution Markov Random Fields and a Two-component Texture Model", Proceedings of SCIA-10, Lappenranta (1997)
<%-- Include the code for the document footer --%> <%@ include file="/arch/footer.jsp" %>