A Local Likelihood Active Contour Model for Medical Image Segmentation
In this thesis, we present a local distribution-based active contour model for medical image segmentation. Our framework uses Bayesian a posterior probability as the driving force for contour evolution. As a generalized form of the Chan-Vese piecewise constant model, our framework relaxes global Gaussian assumptions regarding image pixel intensities, and employs local Gaussians/means as the regional representatives to better account for local intensity variations in images. A global-to-local consistency constraint is enforced over the entire image domain to guarantee the validity of the segmentation results. In addition, available spatial distribution priors can be seamlessly integrated into the segmentation procedure to lead to more meaningful segmentation results. We conduct experiments on both synthetic and real medical images. Comparisons are carried out with Chan-Vese piecewise constant and the Chan-Vese piecewise smooth model, which are the state-of-the-art solution for image segmentation. Experimental results demonstrate the improvements made by our framework.
School Location:USA - Ohio
Source Type:Master's Thesis
Keywords:bayesian posterior probability chan vese gaussian assumptions
Date of Publication:01/01/2007