Metaheuristics for model-based object segmentation

Deformable Models (DMs) try to adapt a curve to maximize its overlap with the contour of an object of interest within an image. These computer vision techniques, extensively used in segmentation problems, require the definition of an optimization method able to deal with noisy and highly-multimodal search spaces, the selection of the algorithm's parameters and a suitable representation for encoding useful information (such as prior shape knowledge), and the initialization of the curve in a location which favors the DM convergence onto the contour of the object of interest. 
This talk will be focused on how these problems can be tackled by applying Metaheuristics: a family of global stochastic techniques designed to solve complex optimization and machine learning problems.