Vectorization of a statistical segmentation
Résumé
We propose an efficient vectorial implementation of a region merging segmentation algorithm. In this algorithm the merging order is based on edge value, and the merging predicate exploits recent statistical investigations. A notable acceleration is obtained by exploiting two forms of parallelism, firstly the Data Level Parallelism by processing edges of the same weight in parallel, secondly the Instruction Level Parallelism. Moreover, the classical UNION-FIND data structure is improved by using local registers to reduce the access time of FIND operations. Finally the implementation could be easily tuned to extract textures (object analysis) or all edges (image enhancement).
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...