The combination of deformable and probabilistic models for image segmentation
C. Ting, D. Metaxes, G. Herman, Samuel Matej · 2003
This paper proposes a new methodology for image segmentation based on the integration of deformable models and Bayesian probabilistic models with Gibbs priors. For each step of the iterative algorithm involving recursive iteration, we first obtain a pixel-based (local) solution based on the Bayesian approach using Gibbs prior, and then use a deformable model to obtain a global solution which will improve the image quality. This combination allows us to overcome the limitations of either of these types of approaches used alone and obtain improved segmentation results independent of the imaging modality used (MRI, CT, ultrasonic image, etc). Another advantage of the method is that the Bayesian approach will filter out the noise from the actual features of the image to increase the SNR during the pixel-based segmentation process.