The watershed transformation is a mid-level operation used in morphological image segmentation. It is well known that watershed transformations are made of very large amount of data and that many operations per pixel must be performed in order to achi...
The watershed transformation is a mid-level operation used in morphological image segmentation. It is well known that watershed transformations are made of very large amount of data and that many operations per pixel must be performed in order to achieve useful result of segmentation.
This thesis proposes a parallel image processing system for watershed algorithm, which is based on Park's multi-access memory system. The proposed parallel image processing system use SIMD computer together with MIMD computer architecture for making to parallelize watersheds.
The proposed parallel image processing system consists of microprocessor, PE modules, DMA controller and sixty seven multi-access memory systems. The microprocessor controls DMA controller, each PE modules, and establish the communication among the host computer and the PE module. The multi-access memory system consists of four PEs and five memory modules.
The function and performance of the proposed architecture is simulated and verified by Cadence Verilog-XL, and we shows that the matrix operation such as comparision with pixels can be effectively implemented with the proposed architecture. Experimental result of proposed system is compared to result of serial-processing systme.