Smooth contour coding with minimal description length active grid segmentation techniques

We analyze the influence of the contour coding term in segmentation techniques based on active grids and on the minimum description length (MDL) principle. These segmentation techniques have been developed up to now with a contour coding term adapted to polygonal objects. However, this approach can lead to degraded segmentation results for smooth contours of objects which can be observed for example in geoscience, medicine or microscopy. We demonstrate that an appropriate choice of the contour coding term can improve segmentation results with MDL active grid approaches in the presence of regions with smooth boundaries. This improvement opens a large class of application domains and still allows one to obtain low computational time.

Research highlights
► Proposition of a MDL active grid segmentation technique adapted to the segmentation of objects with smooth contours.
► This technique can be applied to a large number of noise models for which a polygonal contour description was previously adopted.
► This technique still leads to low computational times and does not require the tuning of any parameter in the optimized criterion.