In our project, we want to analyse the human silhouette. To well understand it, it is better to know each body part. But how to segment the silhouette to those meaningful body parts?
1. One way, we find that someone use Generalized Voronoi Diagram Decompostion to split the silhouette into distinct regions. http://www.mpi-inf.mpg.de/~theobalt/VisualHullTracking/index.html
Using this method, some small regions should be post-processed to form a meaningful human body part.