Title :
Parallel BFS graph traversal on images using structured grid
Author :
Su, Bor-Yiing ; Brutch, Tasneem G. ; Keutzer, Kurt
Author_Institution :
EECS Dept., Univ. of California, Berkeley, CA, USA
Abstract :
Graph algorithms are widely used in image processing techniques. With technology advancements, image sizes are increasing, and the contents inside images are becoming more complex, resulting in increased runtimes for graph algorithms on these images. Breadth First Search (BFS) is a fundamental graph traversal approach. A key to parallelizing graph algorithms used in image processing is to parallelize the BFS graph traversal operation. In this paper, we propose using highly parallelizable structured grid computations to realize the BFS graph traversal operations. This mapping enables efficient implementation of the BFS graph traversal operations on highly parallel manycore platforms. By using such a mapping, we were able to achieve performance gains of 2× to 33× depending on image complexity.
Keywords :
graph theory; image processing; tree searching; BFS graph traversal operation; breadth first search; graph algorithms; image processing; parallel BFS graph traversal; parallelizing graph; structured grid computation; Image segmentation; Magnetic resonance imaging; Partitioning algorithms; Pixel; Runtime; Scalability; Graph Theory; Image Processing; Parallel Programming;
Conference_Titel :
Image Processing (ICIP), 2010 17th IEEE International Conference on
Conference_Location :
Hong Kong
Print_ISBN :
978-1-4244-7992-4
Electronic_ISBN :
1522-4880
DOI :
10.1109/ICIP.2010.5652307