Segmentation and Scene Description:

Segmentation

Recently, we have proposed a segmentation algorithm which combines region and contour imformation. The method begins by detecting the main contours of the scene following the edge extraction algorithm exposed by Cufi et al in 1996. The contours obtained represent the boundaries between regions, so every contour theoretically separates two adjacent regions in the image. Then a set of growing centres are chosen at each side and along the contour. Here, we suppose the seeds belonging to a determined side of the contour are associated to the same region. Then, these seeds are used as samples of the corresponding regions and analysed in the chromatic space. The idea consists on to know a priori some characteristics of regions in order to establish appropriated criteria for the posterior growing processes. This allows to adjust the homogeneity criterion to the region's characteristics using specific criteria on each region. Finally, the seeds simultaneously start a concurrent growth using the criterion established for each region which is based on clustering analysis and convex hull construction.

Related papers