Unsupervised segmentation based on Von Mises circular distributions for orientation estimation in textured images
Résumé
This paper deals with textured images and more particularly with directional textures. We propose a new parametric technique to estimate the orientation field of textures. It consists in partitioning the image into regions with homogeneous orientations, and then to estimate the orientation inside each of these regions, which allows us to maximize the size of the samples used to estimate the orientation without being corrupted by the presence of frontiers between regions. Once estimated the local - hence noisy - orientations of the texture using small filters (3 x 3 pixels), image partitioning is based on the minimization of the stochastic complexity (Minimum Description Length principle) of the orientation field. The orientation fluctuations are modeled with Von Mises probability density functions, leading to a fast and unsupervised partitioning algorithm. The accuracy of the orientations estimated with the proposed method is the compared with other approaches on synthetic images. An application to the processing of real images is finally addressed.