Visual Saliency Model on Multi-GPU

Anis Rahman (1) , Dominique Houzet (1) , Denis Pellerin (1)

1 GIPSA-AGPIG - GIPSA - Architecture, Géométrie, Perception, Images, Gestes

Visual attention models translate the capability of human vision to concentrate only on smaller regions of the visual scene. More precisely, such regions are the spotlight of focus, which either may be an object or a portion of the scene. A number of modalities are used to locate regions of attention like intensity, color, orientation, motion, and many others. The attention model acts as an informationprocessing bottleneck to reduce the overall information into a region of useful information. When guided by salient stimuli, this model falls into a category of a bottom-up approach, which is fast and primitive. On the other hand, models driven by cognition using variable selection criteria are the basis for top-down approaches, and they are slower and more complex. The human visual system uses either a saliency-based or a top-down approach, or the combination of both these approaches to find the spotlight of focus.

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

Anis Rahman, Dominique Houzet, Denis Pellerin. Visual Saliency Model on Multi-GPU. GPU Computing Gems Emerald Edition, Elsevier, pp.451-472, 2011, 978-0-12-384988-5. ⟨10.1016/B978-0-12-384988-5.00030-9⟩. ⟨hal-00595901⟩