基于GPU的多尺度Retinex图像增强算法实现

doi:10.3976/j.issn.1002-4026.2017.03.018

山东科学 ›› 2017, Vol. 30 ›› Issue (3): 103-109.doi: 10.3976/j.issn.1002-4026.2017.03.018

基于GPU的多尺度Retinex图像增强算法实现

李辉，解维浩，刘寿生，盖颖颖

山东省科学院海洋仪器仪表研究所，山东青岛 266001

收稿日期:2016-08-05 出版日期:2017-06-20 发布日期:2017-06-20
作者简介:李辉（1987—），女，硕士，研究方向为图形图像处理、高性能计算。E-mail：lihuihuidou@163.com
基金资助:
山东省科学院青年科学基金（2014QN032）

Realization of multiscale Retinex image enhancement algorithm based on GPU

LI Hui, XIE Wei-hao, LIU Shou-sheng, GAI Ying-ying

Institute of Oceanographic Instrumentation, Shandong Academy of Science, Qingdao 266001,China

Received:2016-08-05 Online:2017-06-20 Published:2017-06-20

摘要/Abstract

摘要：

为提高多尺度Retinex算法的实时性，本文提出了基于GPU的多尺度Retinex图像增强算法，通过对算法进行数据分析和并行性挖掘，将高斯滤波、卷积和对数差分等计算量非常耗时的模块放到GPU中，利用大规模并行线程处理来提高效率。在GeForce GTX 480和CUDA 5.5中进行实验，结果表明该算法能显著提高计算速度，且随着图像分辨率的增加，最大加速比达160倍。

关键词: 多尺度Retinex, GPU, CUDA, 图像增强, 并行计算

Abstract:

To improve the realtime performance of the multiscale Retinex algorithm, a GPU based multiscale Retinex image enhancement algorithm was proposed in this paper. Through the data analysis and parallel mining of the algorithm, timeconsuming modules of the calculation, such as Gauss filter, convolution, and logarithm difference, were implemented in GPU, and the efficiency was improved by using massively parallel processing threads. Experiments were conducted in GeForce GTX 480 and CUDA5.5, and the results showed that the proposed algorithm could significantly improve the computing speed, and with the increasing of the image resolution, the maximum speed up ratio could reach 160 times.

Key words: multi-scale Retinex, parallel computing, GPU, image enhancement, CUDA

中图分类号:

TP391.41

李辉，解维浩，刘寿生，盖颖颖. 基于GPU的多尺度Retinex图像增强算法实现[J]. 山东科学, 2017, 30(3): 103-109.

LI Hui, XIE Wei-hao, LIU Shou-sheng, GAI Ying-ying. Realization of multiscale Retinex image enhancement algorithm based on GPU[J]. SHANDONG SCIENCE, 2017, 30(3): 103-109.

参考文献

［1］尹士畅,喻松林.基于小波变换和直方图均衡的红外图像增强［J］.激光与红外,2013,43(2):225228.
［2］占必超,吴一全,纪守新.基于平稳小波变换和Retinex的红外图像增强方法［J］.光学学报，2010,30(10):27882793.
［3］李庆忠,刘清.基于小波变换的低照度图像自适应增强算法［J］.中国激光,2015,42(2):0209001.
［4］WANG Y K, HUANG W B. A CUDAenabled parallel algorithm for accelerating retinex ［J］. Journal of RealTime Image Processing, 2014, 9 (3): 407425.
［5］LAND E H. Recent advances in Retinex theory ［M］// Central and Peripheral Mechanisms of Colour Vision. UK : Palgrave Macmillan,1985: 517.
［6］刘瑞剑,低能见度条件下图像清晰化处理研究［D］.太原: 中北大学, 2008.
［7］ LEE C H, SHIH J L, LIEN C C, et al. Adaptive multiscale retinex for image contrast enhancement ［M］// Proceedings of the 2013 International Conference on SignalImage Technology & InternetBased Systems. Washington, DC, USA： IEEE Computer Society, 2013: 4350.
［8］NVIDIA .CUDA ［EB/OL］. ［20160403］. http://www.nvidia.cn/object/cudacn.html.
［9］徐元旭, SIMT线程调度模型分析及优化［D］.哈尔滨: 哈尔滨工业大学, 2013.
［10］COOK S.CUDA并行程序设计:GPU编程指南［M］.北京: 机械工业出版社,2014: 191200.
［11］ JIAN L H,WANG C,LIU Y, et al. Parallel data mining techniques on Graphics Processing Unit with Compute Unified Device Architecture (CUDA) ［J］. The Journal of Supercomputing, 2013, 64 (3): 942967.
［12］ YANG Z Y, ZHU Y T, PU Y. Parallel Image Processing Based on CUDA ［M］//Proceedings of the 2008 International Conference on Computer Science and Software Engineering. Washington, DC, USA: IEEE Computer Society, 2008: 198201.
［13］NVIDA. Geforce GTX 480 Computing Processor Board ［EB/OL］. ［2016-04-03］. http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-480/specifications.

[1]	李向春, 张浩, 刘晓燕, 宗芳伊, 刘军礼. 基于透射率优化和颜色修正的水下图像增强方法[J]. 山东科学, 2019, 32(2): 7-14.
[2]	姜鹏. 基于Hadoop的即时查询分析平台[J]. 山东科学, 2015, 28(5): 115-119.
[3]	林云，张少龙，张庆刚，张怿慈. 蛋白质微秒级折叠过程的快速模拟[J]. 山东科学, 2015, 28(3): 96-103.
[4]	王加磊, 孟现美, 张庆刚 ,张怿慈, 张少龙. HIV-1 蛋白酶异位抑制剂体系的长时间分子动力学模拟[J]. J4, 2013, 26(2): 1-6.
[5]	张庆科, 杨波, 王琳, 陈贞翔. 基于异构模式的云计算关键技术研究[J]. J4, 2011, 24(5): 56-62.

基于GPU的多尺度Retinex图像增强算法实现

Realization of multiscale Retinex image enhancement algorithm based on GPU

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 5

Metrics

本文评价

推荐阅读 0