Developing Bayesian methods for biomedical image processing

• Main Author: Zhang, Huaizhong
• Published: University of Ulster 2010
• Subjects: 616.0704
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.553883

The broad objective of this thesis is the development of application-specific, computation ally efficient image processing approaches in the light of Bayesian theory when applied to Biomedical Imagery. Due to its rigorous theoretical background and precise formula expression, the Bayesian method has been successfully applied in image processing over the years. Based on the investigation of relationships between various optimization criteria within the Bayesian framework, the main contributions of this thesis lie in the development of four different, but interrelated, image processing approaches with strong connections to curve evolution and estimation techniques. The first contribution of this thesis is the development of a novel adaptive bias correction approach by applying the Markov Chain Monte Carlo technique in image alignment. A hierarchical model is proposed for estimating the accumulated bias and the unknown mean is simulated by a Gibbs sampler. Then, the scale parameters are estimated such that the equalization transformation can be performed semi-automatically. Another contribution of this thesis is the development of an application-specific improvement to the classical Geodesic Active Contours method where a prior model is used to incorporate prior information into the scheme. The third contribution of this thesis is the development of a coupling method in boundary finding of Regions of Interest, which combines advantages of both edge-based and region-based techniques. The key behind this approach is the use of a mixture modelling technique that produces the image energy by applying the Bayesian method in order to control curve evolution. The final contribution of this thesis is the development of a termination criterion that offers a reliable method to control curve evolution and help curve convergence. Here we apply the Bayesian method to generate a stability index which is used to control curve evolution leading to more stable and efficient curve evolution.