The GPU as a co-processor for volume image segmentation
Dipl.-Inform. Björn Wagner
July 24, 2009, 8 a.m. BA 9912Volume image processing is a very data and computation intensive task. Especially complex segmentation operations, like separation of touching objects or cell reconstruction of open foam cells, are usually very expensive.
Due to graphic intense computer games, graphic processing units have emerged impressing computation power during the last years. With the improved programmability of graphic processing units, through technologies like Nvidia Cuda and BrookGPU, a cheap and powerful high performance computing platform is now available.
In this talk I will present a novel approach for a parallel watershed transformation, which is a powerful method of mathematical morphology for gray-scale image segmentation. I will discuss its application as a key algorithm of the cell reconstruction tool-chain with focus on the implementation for the gpu computing platform compared with the implementation on the multicore-platform.