The Implementation of Parallelization for The Time-Lapse Fusion Algorithm

碩士 === 輔仁大學 === 電機工程學系碩士班 === 104 === Time-Lapse Fusion(TLF) is an image sequence blending algorithm which can create high dynamic range image sequences. Through a temporally fading mechanism, the algorithm generates a long exposure effect. However, the computational loading of TLF is tremendous due...

Full description

Bibliographic Details
Main Authors: Yuan-Chung Tai, 戴元駿
Other Authors: Yuan-Kai Wang
Format: Others
Language:zh-TW
Published: 2016
Online Access:http://ndltd.ncl.edu.tw/handle/03819176528368540277
id ndltd-TW-104FJU00428013
record_format oai_dc
spelling ndltd-TW-104FJU004280132016-11-18T04:43:22Z http://ndltd.ncl.edu.tw/handle/03819176528368540277 The Implementation of Parallelization for The Time-Lapse Fusion Algorithm 縮時融合演算法平行化之實現 Yuan-Chung Tai 戴元駿 碩士 輔仁大學 電機工程學系碩士班 104 Time-Lapse Fusion(TLF) is an image sequence blending algorithm which can create high dynamic range image sequences. Through a temporally fading mechanism, the algorithm generates a long exposure effect. However, the computational loading of TLF is tremendous due to its repeated computational pattern with a single program multiple data (SPMD) fashion. In this paper we devise a parallel TLF method called GTLF using GPU and OpenCL to parallelize the serial TLF. The GTLF uses OpenCL kernel under the GPU architecture and improves the efficiency by memory hierarchy. A computational analysis of the GTLF is discussed in details in this paper. We compare the run time and speedup between the GTLF and the serial TLF called CTLF. The experiments of this paper are conducted on the GPU TeslaC2050 and the OpenCL 1.1 in OpenCV OCL module. Experimental results show that the GTLF can achieve 71 times speedup compared with the optimized single-threaded CTLF in image resolution 4096 x 4096, and parallel efficiency with 54%. Our experimental results demonstrate that using GPU and OpenCL can greatly improve the performance of the TLF algorithm. Yuan-Kai Wang 王元凱 2016 學位論文 ; thesis 78 zh-TW
collection NDLTD
language zh-TW
format Others
sources NDLTD
description 碩士 === 輔仁大學 === 電機工程學系碩士班 === 104 === Time-Lapse Fusion(TLF) is an image sequence blending algorithm which can create high dynamic range image sequences. Through a temporally fading mechanism, the algorithm generates a long exposure effect. However, the computational loading of TLF is tremendous due to its repeated computational pattern with a single program multiple data (SPMD) fashion. In this paper we devise a parallel TLF method called GTLF using GPU and OpenCL to parallelize the serial TLF. The GTLF uses OpenCL kernel under the GPU architecture and improves the efficiency by memory hierarchy. A computational analysis of the GTLF is discussed in details in this paper. We compare the run time and speedup between the GTLF and the serial TLF called CTLF. The experiments of this paper are conducted on the GPU TeslaC2050 and the OpenCL 1.1 in OpenCV OCL module. Experimental results show that the GTLF can achieve 71 times speedup compared with the optimized single-threaded CTLF in image resolution 4096 x 4096, and parallel efficiency with 54%. Our experimental results demonstrate that using GPU and OpenCL can greatly improve the performance of the TLF algorithm.
author2 Yuan-Kai Wang
author_facet Yuan-Kai Wang
Yuan-Chung Tai
戴元駿
author Yuan-Chung Tai
戴元駿
spellingShingle Yuan-Chung Tai
戴元駿
The Implementation of Parallelization for The Time-Lapse Fusion Algorithm
author_sort Yuan-Chung Tai
title The Implementation of Parallelization for The Time-Lapse Fusion Algorithm
title_short The Implementation of Parallelization for The Time-Lapse Fusion Algorithm
title_full The Implementation of Parallelization for The Time-Lapse Fusion Algorithm
title_fullStr The Implementation of Parallelization for The Time-Lapse Fusion Algorithm
title_full_unstemmed The Implementation of Parallelization for The Time-Lapse Fusion Algorithm
title_sort implementation of parallelization for the time-lapse fusion algorithm
publishDate 2016
url http://ndltd.ncl.edu.tw/handle/03819176528368540277
work_keys_str_mv AT yuanchungtai theimplementationofparallelizationforthetimelapsefusionalgorithm
AT dàiyuánjùn theimplementationofparallelizationforthetimelapsefusionalgorithm
AT yuanchungtai suōshírónghéyǎnsuànfǎpíngxínghuàzhīshíxiàn
AT dàiyuánjùn suōshírónghéyǎnsuànfǎpíngxínghuàzhīshíxiàn
AT yuanchungtai implementationofparallelizationforthetimelapsefusionalgorithm
AT dàiyuánjùn implementationofparallelizationforthetimelapsefusionalgorithm
_version_ 1718394542027177984