Implementation of a Bayesian Statistics Segmentation Algorithm in GPUs

FABRICIO FIORETTI; DAVID LEMMA; MILLÁN, EMMANUEL NICOLÁS; ROBERTO ISOARDI; EDUARDO M. BRINGA

Buenos Aires

Simposio; V Latin American Symposium on High Performance Computing HPCLatAm 2012; 2012

Facultad de Ciencias Exactas y Naturales Universidad de Buenos Aires

The increasing interest in the implementation of scientific algorithms in (Graphic Processing Units) GPUs stems from the high levels of computing acceleration which can be obtained through their use, for instance in image processing. Tomographic image segmentation using a Bayesian approach relies on the statistical inference of global parameters unsuitable for local data treatment, which is the strength of the highly parallel processing architecture of GPUs. We solve this problem with a paralel algorithm, and obtain a perfomance which is up to 57x of the serial CPU version of the code.