基于图形处理器的高速中值滤波算法

doi:10.13229/j.cnki.jdxbgxb20180050

吉林大学学报(工学版) ›› 2019, Vol. 49 ›› Issue (3): 979-985.doi: 10.13229/j.cnki.jdxbgxb20180050

基于图形处理器的高速中值滤波算法

托乎提努尔^1,²(),张海龙^1,³(),王杰¹,王娜^1,³,冶鑫晨¹,王万琼¹

1. 中国科学院新疆天文台，乌鲁木齐 830011
2. 中国科学院大学，北京 100049
3. 中国科学院射电天文重点实验室，南京 210008

收稿日期:2018-01-16 出版日期:2019-05-01 发布日期:2019-07-12
通讯作者: 张海龙 E-mail:nuer@xao.ac.cn;zhanghailong@xao.ac.cn
作者简介:托乎提努尔(1986?)，男，博士研究生. 研究方向：GPU并行计算，天文信号处理. E?mail:nuer@xao.ac.cn
基金资助:
“973”国家重点基础研究发展计划项目(2015CB857100)

High speed median filtering algorithm based on graphics processing unit

Tohtonur^1,²(),Hai⁃long ZHANG^1,³(),Jie WANG¹,Na WANG^1,³,Xin⁃chen YE¹,Wan⁃qiong WANG¹

1. Xinjiang Astronomical Observatory, Chinese Academy of Sciences, Urumqi 830011, China
2. University of Chinese Academy of Sciences, Beijing 100049, China
3. Key Laboratory of Radio Astronomy, Chinese Academy of Sciences, Nanjing 210008, China

Received:2018-01-16 Online:2019-05-01 Published:2019-07-12
Contact: Hai?long ZHANG E-mail:nuer@xao.ac.cn;zhanghailong@xao.ac.cn

摘要/Abstract

摘要：

针对中央处理器(CPU)平台中值滤波算法在实际应用中运算速率低且实时信号处理性能较差的问题，提出了一种基于图形处理器（GPU）的并行高速中值滤波算法。该算法采用统一计算设备架构（CUDA）并行架构对大规模数据处理进行了优化，从而有效提高了中值滤波算法的计算效率，实现了中值滤波的实时数据处理。通过构建GPU可任意伸缩的动态数组、优化多维索引的线性化方法解决了GPU动态显存空间分配问题。仿真试验结果表明：基于TITAN X GPU的5×5中值滤波，对4096像素×4096像素的图像处理计算速度比CPU平台提高了438倍。在同等计算规模条件下GPU高速中值滤波算法可大大提高计算性能。

关键词: 信息处理技术, 信号处理, 中值滤波, 统一计算设备架构, 图形处理器, 并行算法

Abstract:

Low computational rate and poor performance in real?time signal processing are the main problems for the median filtering algorithm in the practical applications. This paper proposed a high?speed parallel median filtering algorithm based on Graphics Processing Unit (GPU). The algorithm uses Compute Unified Device Architecture (CUDA) to optimize large?scale data processing and it is implemented on NVIDIA GPUs to improved its computational efficiency. The GPU′s dynamic memory space is allocated by constructing GPU?scalable dynamic array and optimization of multidimensional index linearization methods. Experiment results show that, the 5×5 median filtering based on TITAN X GPU is approximately 438x faster than CPU algorithm for processing of 4096×4096 pixel images. The GPU based median filtering can greatly improve the computing performance of algorithm under the same computing conditions.

Key words: information processing technology, signal processing, median filtering, compute unified device architecture(CUDA), graphics processing unit(GPU), parallel algorithm

中图分类号:

TN911.7

托乎提努尔,张海龙,王杰,王娜,冶鑫晨,王万琼. 基于图形处理器的高速中值滤波算法[J]. 吉林大学学报(工学版), 2019, 49(3): 979-985.

Tohtonur,Hai⁃long ZHANG,Jie WANG,Na WANG,Xin⁃chen YE,Wan⁃qiong WANG. High speed median filtering algorithm based on graphics processing unit[J]. Journal of Jilin University(Engineering and Technology Edition), 2019, 49(3): 979-985.

图/表 12

图1

图2

图3

图4

图5

图6

表1

图7

图8

表2

表3

图9

参考文献 12

1	WoźniakM, PołapD, GabryelM, et al. Can we process 2D images using artificial bee colony?[C]∥International Conference on Artificial Intelligence and Soft Computing,Springer,Cham,2015:660⁃671.
2	赵海英,张小利,李雄飞,等. 基于多尺度Meanshift图像去噪算法[J]. 吉林大学学报:工学版,2014,44(5):1417⁃1422.
	ZhaoHai⁃ying, ZhangXiao⁃li, LiXiong⁃fei, et al. Image denoising algorithom basid on multi⁃scale Meanshift[J]. Journal of Jilin University (Engineering and Technology Edition),2014,44(5):1417⁃1422.
3	许景波,袁怡宝,崔晓萌,等. 表面测量中高斯滤波中线的有理逼近实现[J]. 吉林大学学报:工学版,2014,44(5):1347⁃1352.
	XuJing⁃bo, YuanYi⁃bao, CuiXiao⁃meng, et al. Rational approximation implementation approach to determine Gaussian filtering mean line in surface roughness measurement[J]. Journal of Jilin University(Engineering and Technology Edition),2014,44(5):1347⁃1352.
4	JuholaM, KatajainenJ,RaitaT. Comparison of algorithms for standard median filtering[J]. IEEE Transactions on Signal Processing,1991,39(1):204⁃208.
5	TomasiC, ManduchiR. Bilateral filtering for gray and color images[C]∥Sixth International Conference on Computer Vision,Bombay,India,1998:839⁃846.
6	JwoD J, WangS H. Adaptive fuzzy strong tracking extended Kalman filtering for GPS navigation[J]. IEEE Sensors Journal,2007,7(5):778⁃789.
7	王宇新,贺圆圆. 基于FPGA的快速中值滤波算法[J]．计算机应用研究,2009,26(1):224⁃226.
	WangYu⁃xin, HeYuan⁃yuan. FPGA⁃based algorithm of fast median filter[J]. Application Research of Computers,2009,26(1):224⁃226.
8	OwensJ D, HoustonM, LuebkeD, et al. GPU computing[J]. Proceedings of the IEEE,2008,96(5):879⁃899.
9	RankaS, SahniS. Efficient serial and parallel algorithms for median filtering[J]. IEEE Transactions on Signal Processing,1991,39(6):1462⁃1466.
10	MichaelGarland,GrandS L,NickollsJ,et al.Parallel computing experiences with CUDA[J]. IEEE Micro,2008,28(4):13⁃27.
11	RyooS, RodriguesC I, BaghsorkhiS S, et al. Optimization principles and application performance evaluation of a multithreaded GPU using CUDA[C]∥Proceedings of the 13th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming,Salt Lake City,Utah,USA,2008:73⁃82.
12	BattiatoP S. High performance median filtering algorithm based on NVIDIA GPU computing[J/OL].[2018⁃01⁃04].http:∥ceur⁃ws.org/Vol⁃1543/p1.pdf.

相关文章 15

[1]	郭立民,陈鑫,陈涛. 基于AlexNet模型的雷达信号调制类型识别[J]. 吉林大学学报(工学版), 2019, 49(3): 1000-1008.
[2]	刘鲁涛,李娜. 基于互质阵列的信源数估计[J]. 吉林大学学报(工学版), 2019, 49(3): 986-993.
[3]	付银娟,李勇,徐丽琴,张昆辉. NLFM⁃Costas射频隐身雷达信号设计及分析[J]. 吉林大学学报(工学版), 2019, 49(3): 994-999.
[4]	苏寒松,代志涛,刘高华,张倩芳. 结合吸收Markov链和流行排序的显著性区域检测[J]. 吉林大学学报(工学版), 2018, 48(6): 1887-1894.
[5]	徐岩,孙美双. 基于卷积神经网络的水下图像增强方法[J]. 吉林大学学报(工学版), 2018, 48(6): 1895-1903.
[6]	李居朋,张祖成,李墨羽,缪德芳. 基于Kalman滤波的电容屏触控轨迹平滑算法[J]. 吉林大学学报(工学版), 2018, 48(6): 1910-1916.
[7]	黄勇,杨德运,乔赛,慕振国. 高分辨合成孔径雷达图像的耦合传统恒虚警目标检测[J]. 吉林大学学报(工学版), 2018, 48(6): 1904-1909.
[8]	应欢,刘松华,唐博文,韩丽芳,周亮. 基于自适应释放策略的低开销确定性重放方法[J]. 吉林大学学报(工学版), 2018, 48(6): 1917-1924.
[9]	陆智俊,钟超,吴敬玉. 星载合成孔径雷达图像小特征的准确分割方法[J]. 吉林大学学报(工学版), 2018, 48(6): 1925-1930.
[10]	刘仲民,王阳,李战明,胡文瑾. 基于简单线性迭代聚类和快速最近邻区域合并的图像分割算法[J]. 吉林大学学报(工学版), 2018, 48(6): 1931-1937.
[11]	单泽彪,刘小松,史红伟,王春阳,石要武. 动态压缩感知波达方向跟踪算法[J]. 吉林大学学报(工学版), 2018, 48(6): 1938-1944.
[12]	姚海洋, 王海燕, 张之琛, 申晓红. 双Duffing振子逆向联合信号检测模型[J]. 吉林大学学报(工学版), 2018, 48(4): 1282-1290.
[13]	全薇, 郝晓明, 孙雅东, 柏葆华, 王禹亭. 基于实际眼结构的个性化投影式头盔物镜研制[J]. 吉林大学学报(工学版), 2018, 48(4): 1291-1297.
[14]	陈涛, 崔岳寒, 郭立民. 适用于单快拍的多重信号分类改进算法[J]. 吉林大学学报(工学版), 2018, 48(3): 952-956.
[15]	陈绵书, 苏越, 桑爱军, 李培鹏. 基于空间矢量模型的图像分类方法[J]. 吉林大学学报(工学版), 2018, 48(3): 943-951.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

Specifications	Xeon(R) E5?2630	GTX TITAN X
Total amount of core	8	3072
Frequency	2.40 GHz	1000 MHz
Global memory	64 GB(RAM)	12 GB
Cache size	20480 kB	3072 kB
Shared memory per block	-	49152 bytes
Warp size	-	32
Maximum threads per block	-	1024
Maximum threads per multiprocessor	-	2048
Max dimension size of a block (x,y,z)	-	(1024, 1024, 64)
Registers available per block	-	65536

Image size	Xeon(R) E5?2630	GTX TITAN X
128×128	5.559	0.3106
256×256	23.0534	0.4264
512×512	99.9552	0.6893
1024×1024	331.881	1.8339
2048×2048	1117.41	4.783
4096×4096	3870.38	15.8201

Image size	Xeon(R) E5?2630	GTX TITAN X
128×128	37.9345	0.4178
256×256	151.973	0.8302
512×512	645.738	2.0857
1024×1024	2281.34	6.6295
2048×2048	8131.87	20.8194
4096×4096	27940.2	63.7018

基于图形处理器的高速中值滤波算法

High speed median filtering algorithm based on graphics processing unit

RICH HTML

PDF (PC)

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 12

相关文章 15

Metrics

本文评价

推荐阅读 10