Stochastic and biological metaphor parameter estimation on the Gaussian mixture model and image segmentation by Markov random field

The model parameters of image in real life applications are usually unknown and are necessary for any image processing such as image segmentation. Parameter estimation, labels, can be done from observed image. We proposed use of probabilistic transition rules based on biological metaphor, Genetic Algorithm (GA), standard Expectation Maximization (EM), Simulated Annealing (SA) and mix of these methods for learning Gaussian mixture components to achieve accurate parameter estimation on images. We also introduced modified implementations of SA for image segmentation. The segmentation procedure is based on Markov random field (MRF) model for describing regions within an image. We proposed a random cost function for computing a posterior energy function in SA. The proposed modified Simulated Annealing (SA-RCF) method depicts more robust performance for image segmentation than standard SA at the same computational cost. Alternatively, we proposed a multi-resolution (MR) approach based on MRF, which offers a robust segmentation for noisy images with significant reduction in the computational cost on phantom images. This thesis proposes accurate and stable solution methods for both parameter estimation and image segmentation for dental images. All proposed methods were evaluated on CT phantom images and applied on μCT images