Segmentation of noisy images using information theory based approaches

In this presentation, we propose to discuss some interesting properties of segmentation techniques based on the minimization of the stochastic complexity. We emphasize the general framework provided by the minimization of the stochastic complexity for segmentation purpose, some of its main advantages and also some of the motivating perspectives that are open by such approaches. We illustrate this presentation with different results obtained in our research group with polygonal parametric shape descriptions, level set models of contours and polygonal grids to partition images into an arbitrary number of homogeneous regions.