Parallel implementation of multi-resolution filter band with novel Adaptive Load Balancing Algorithm on consumer-level multicore systems

The widespread usage of the Discrete Wavelet Transform (DWT) has motivated the development of Parallel Discrete Wavelet Transform (PDWT) on multicore computer systems. Several studies that have implemented PDWT by utilization of parallel technology have shown that it is an effective approach to spee...

Full description

Saved in:
Bibliographic Details
Main Authors: Majid, M. W., Mirzaei, G., Jamali, M. M.
Format: Conference Proceeding
Language:English
Subjects:
Discrete wavelet transforms
Instruction sets
Load management
Multicore processing
Parallel processing
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The widespread usage of the Discrete Wavelet Transform (DWT) has motivated the development of Parallel Discrete Wavelet Transform (PDWT) on multicore computer systems. Several studies that have implemented PDWT by utilization of parallel technology have shown that it is an effective approach to speed up the wavelet transformation by a considerable factor. However, the parallelism is implemented in multi-machine network environment which affects the system maintenance and the bandwidth speed. Therefore, multi-core technology has great importance as it gives an intermediate approach to system speedup without jeopardizing on performance. In this work, wavelet transform is implemented on a single core/multi-core system and developed in C# using. Net framework 4.0. This study presents an efficient approach to compute the DWT transform with Adaptive Load Balancing Algorithm (ALBA) which is applied on standard form on parallel general-purpose computers. This technique does not introduce any restriction on the size of the input data or on the transform parameters. Complete use of the available processor parallelism, modularity, and scalability are achieved. Theoretical and experimental evaluations and comparisons are given with respect to traditional parallelization.
ISSN:1548-3746
1558-3899
DOI:10.1109/MWSCAS.2012.6292128