Threewise: a local variance algorithm for GPU

Florian Gouin (1) , Corinne Ancourt (1, 2) , Christophe Guettier (3)

1 CRI - Centre de Recherche en Informatique
2 Mines Paris - PSL (École nationale supérieure des mines de Paris)
3 SAGEM Défense Sécurité [Massy]

Variance computation is commonly used in many fields like in image processing to improve local contrasts. This article is not only about developing and placing an algorithm of variance computation for graphical processors, it will also introduce its optimisation in terms of precision and computing time in relation to architectural constraints of graphical processors. Our algorithm enables to improve complexity in O(N log N) and brings a speedup of 112 compared to the classical formulation and of 4 regarding the optimized Pairwise algorithm.

Mots clés: Variance; computation; kernel computation; GPU; image processing; local contrats enhancement

Domaines: Traitement des images [eess.IV]; Calcul parallèle, distribué et partagé [cs.DC]; Parallélisation; Performance et fiabilité [cs.PF]

Florian Gouin, Corinne Ancourt, Christophe Guettier. Threewise: a local variance algorithm for GPU. 19th IEEE International Conference on Computational Science and Engineering (CSE 2016), Aug 2016, Paris, France. pp.257-262. ⟨hal-01359482⟩