Parallel image processing on a shared-memory multiprocessor
The objective of the research work associated with this thesis has been to investigate parallel image processing techniques on an M.I.M.D (Multiple Instruction-stream Multiple Data-stream) type parallel computer. A number of commonly used low level image processing algorithms have been implemented utilising two different forms of parallelisms on the multimicroprocessor CYBA-M, which has a direct shared memory (DSM) architecture. Results have shown that,'Image' parallelism, which involves assigning groups of pixels to different processors for concurrent processing, is the more efficient implementation for such image processing operations. To increase processing efficiency by distributing processing load evenly among the processors, static and dynamic load allocation methods have been studied. A dynamic load allocation algorithm used with Image parallelism has been developed which has proved to be efficient, flexible and applicable to both feature-dependent and feature independent operations. It has been shown that by selecting the optimum load-size for a particular image processing operation, overheads can be kept to a minimum, resulting in maximum efficiency.
- Research Organization:
- Manchester Univ. (United Kingdom)
- OSTI ID:
- 5585215
- Country of Publication:
- United States
- Language:
- English
Similar Records
Gaussian techniques on shared memory multiprocessor computers
Distributed memory management system for pasm