面向GPU的直方图统计图像增强并行算法

doi:10.3778/j.issn.1673-9418.2103059

计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (10): 2273-2285.DOI: 10.3778/j.issn.1673-9418.2103059

面向GPU的直方图统计图像增强并行算法

肖汉¹^,³, 孙陆鹏¹, 李彩林²^,⁺(), 周清雷³

1.郑州师范学院信息科学与技术学院,郑州 450044
2.山东理工大学建筑工程学院,山东淄博 255000
3.郑州大学计算机与人工智能学院,郑州 450001

收稿日期:2021-03-17 修回日期:2021-05-25 出版日期:2022-10-01 发布日期:2021-06-23
通讯作者: + E-mail: licailin@sdut.edu.cn
作者简介:肖汉（1970—）,男,湖北武汉人,博士,教授,CCF高级会员,主要研究方向为高性能计算、图像处理、大型并行数值软件。
孙陆鹏（1970—）,男,河南巩义人,硕士,讲师,CCF会员,主要研究方向为并行计算、图像处理。
李彩林（1985—）,男,安徽安庆人,博士,副教授,主要研究方向为数字摄影测量与计算机视觉、数字图像处理。
周清雷（1962—）,男,河南郑州人,博士,教授,博士生导师,CCF杰出会员,主要研究方向为并行算法、图像处理、并行计算。
基金资助:
国家自然科学基金(41601496);国家自然科学基金(61572444);河南省高等学校重点科研项目(22A520049);黄河中下游数字地理技术教育部重点实验室（河南大学）开放基金(GTYR202004);自然资源部大湾区地理环境监测重点实验室（深圳大学）开放基金(SZU51029202003);山东省艺术科学重点课题(ZD202008267);山东省艺术科学重点课题(201806353)

GPU-Oriented Parallel Algorithm for Histogram Statistical Image Enhancement

XIAO Han¹^,³, SUN Lupeng¹, LI Cailin²^,⁺(), ZHOU Qinglei³

1. School of Information Science and Technology, Zhengzhou Normal University, Zhengzhou 450044, China
2. School of Civil and Architectural Engineering, Shandong University of Technology, Zibo, Shandong 255000, China
3. School of Computer and Artificial Intelligence, Zhengzhou University, Zhengzhou 450001, China

Received:2021-03-17 Revised:2021-05-25 Online:2022-10-01 Published:2021-06-23
About author:XIAO Han, born in 1970, Ph.D., professor, senior member of CCF. His research interests include high performance computing, image processing and large-scale parallel numerical software.
SUN Lupeng, born in 1970, M.S., lecturer, member of CCF. His research interests include parallel computing and image processing.
LI Cailin, born in 1985, Ph.D., associate professor. His research interests include digital photogrammetry and computer vision and digital image processing.
ZHOU Qinglei, born in 1962, Ph.D., professor, Ph.D. supervisor, distinguished member of CCF. His research interests include parallel algorithm, image processing and parallel computing.
Supported by:
National Natural Science Foundation of China(41601496);National Natural Science Foundation of China(61572444);Key Scientific Research Projects of Colleges and Universities of Henan Province(22A520049);Open Fund of Key Laboratory of Geospatial Technology for the Middle and Lower Yellow River Regions (Henan University) Ministry of Education(GTYR202004);Open Foundation of the Key Laboratory of Geo-Environmental Monitoring of Great Bay Area (Shenzhen University) the Ministry of Natural Resources of China(SZU51029202003);Key Project of Art Science in Shandong Province(ZD202008267);Key Project of Art Science in Shandong Province(201806353)

摘要/Abstract

摘要：

直方图统计在图像增强和目标检测等领域有着重要的应用。然而,随着图像规模不断增大、实时性要求越来越高,直方图统计局部增强算法的处理过程较慢,达不到预期满意的速度。针对这一不足,在图形处理器（GPU）平台上实现了直方图统计图像增强算法的并行处理,提升了处理大幅面数字图像的处理速度。首先,通过充分利用统一计算设备架构（CUDA）活动线程块和活动线程来并行处理不同的子图像块和像素点,提升了数据访问的效率。然后,采用内核配置参数优化和数据并行计算技术,实现了直方图统计图像增强算法在GPU平台上的并行化。最后,采用主机端和设备端间高效的数据传输模式,进一步缩短了系统在异构计算平台上的执行时间。研究表明,对于像幅大小不同的图像,图像直方图统计并行算法的处理速度相比CPU串行算法均有两个数量级的提高,处理一幅像幅大小为3 241×3 685的图像需要787.11 ms,并行算法的处理速度提高了261.35倍,为实现实时大规模图像处理奠定了良好基础。

关键词: 直方图统计, 局部增强, 局部均值, 图形处理器（GPU）, 统一计算设备架构（CUDA）, 并行算法

Abstract:

Histogram statistics has important applications in the fields of image enhancement and target detection. However, with the increasing size of the image and the higher real-time requirements, the processing process of the histogram statistical local enhancement algorithm is slow and cannot reach the expected satisfactory speed. In view of this deficiency, this paper realizes the parallel processing of histogram statistical image enhancement algorithm on graphics processing unit (GPU) platform, which improves the processing speed of large format digital images. Firstly, the efficiency of data access is improved by making full use of compute unified device architecture (CUDA) active thread block and active thread to process different sub-image blocks and pixels in parallel. Then, the paralle-lization of histogram statistical image enhancement algorithm on GPU platform is realized by using kernel configu-ration parameter optimization and data parallel computing technology. Finally, the efficient data transmission mode between the host and the device is adopted, which further shortens the execution time of the system on the hetero-geneous computing platform. The results show that for images with different image sizes, the processing speed of the image histogram statistical parallel algorithm is two orders of magnitude higher than that of the CPU serial algorithm. It takes 787.11 ms to process an image with an image size of 3241×3685. The processing speed of the parallel algo-rithm is increased by 261.35 times. It lays a good foundation for the realization of real-time large-scale image processing.

Key words: histogram statistics, local enhancement, local mean, graphics processing unit (GPU), compute unified device architecture (CUDA), parallel algorithm

中图分类号:

TP311

肖汉, 孙陆鹏, 李彩林, 周清雷. 面向GPU的直方图统计图像增强并行算法[J]. 计算机科学与探索, 2022, 16(10): 2273-2285.

XIAO Han, SUN Lupeng, LI Cailin, ZHOU Qinglei. GPU-Oriented Parallel Algorithm for Histogram Statistical Image Enhancement[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(10): 2273-2285.

图/表 13

图1 图像拓展示意图

Fig.1 Schematic diagram of image expansion

表1 直方图统计算法主要功能运行时间占比

Table 1 Proportion of running time of main functions of histogram statistical algorithm

主要功能	执行时间/ms	执行时间占比/%
初始化直方图统计图像	36.00	0.032
灰度级概率估计计算	1 212.00	1.080
图像均值求解	10.00	0.009
图像方差求解	14.00	0.012
图像扩充	53.00	0.047
局部直方图统计计算	110 193.00	97.960
其他部分	969.00	0.860
总计	112 487.00	100.000

图2 局部直方图统计的并行性示意图

Fig.2 Parallelism diagram of local histogram statistics

图3 直方图统计并行算法执行模式

Fig.3 Execution mode of histogram statistical parallel algorithm

图4 直方图统计划分

Fig.4 Histogram statistical segmentation

表2 线程块维度对运算速度的影响

Table 2 Influence of thread block dimension on computing speed

线程块中线程数	运算时间/ms
8×8	1 572.53
16×16	1 554.41
24×24	1 593.84
32×32	1 882.17

表3 实验环境配置

Table 3 Experimental environment configuration

参数	设置
CPU	Intel Core i5-7400（四核心）, 主频3.2 GHz
内存	8 GB
显卡	NVIDIA GTX 1070
显示存储器	8 GB GDDR5
CUDA核数	1 920
每个SM的最大激活线程数	2 048
线程块中的线程数	1 024
线程块维度	1 024×1 024×64
网格维度	(2³¹-1)×65 535×65 535

图5 直方图统计图像增强效果

Fig.5 Histogram statistics image enhancement effect

表4 直方图统计算法运行时间对比

Table 4 Comparison of running time of histogram statistical algorithm

像幅大小	串行处理时间/ms	并行处理时间/ms
像幅大小	串行处理时间/ms	OpenMP	CUDA
547×754	7 494.00	5 677.27	40.23
1 246×1 652	38 338.00	21 538.20	156.49
2 265×2 768	112 487.00	52 564.02	432.14
3 241×3 685	205 714.00	84 655.96	787.11
4 357×4 872	370 229.00	120 204.22	1 554.41
5 361×5 763	592 134.00	170 643.80	2 556.55

表5 直方图统计算法性能对比

Table 5 Performance comparison of histogram statistical algorithm

像幅大小	加速比
像幅大小	OpenMP	CUDA
547×754	1.32	186.28
1 246×1 652	1.78	244.98
2 265×2 768	2.14	260.30
3 241×3 685	2.43	261.35
4 357×4 872	3.08	238.18
5 361×5 763	3.47	231.61

图6 串/并行算法直方图对比

Fig.6 Histogram comparison of series/parallel algorithm

图7 不同架构下直方图统计算法运算时间对比

Fig.7 Comparison of running time of histogram statistical algorithm under different architectures

图8 直方图统计并行算法加速比趋势

Fig.8 Speedup trend of histogram statistical parallel algorithm

参考文献 33

[1]	CHEN X, ZHANG Y, LIN L, et al. Efficient anti-glare ceramic decals defect detection by incorporating homomorphic filte-ring[J]. Computer Systems Science and Engineering, 2021, 36(3): 551-564. DOI URL
[2]	HAMIDREZA S R, MOJTABA S, ANAHITA F K, et al. Apparent diffusion coefficient (ADC) and first-order histogram statistics in differentiating malignant versus benign menin-gioma in adults[J]. Iranian Journal of Radiology, 2019, 16(1): 1-9.
[3]	ADAMSKI M, SARNACKI K, SAEED K, et al. Binary hand-writing image enhancement by directional field-guided morp-hology[J]. Information Sciences, 2021, 551(11): 168-183. DOI URL
[4]	ZHAO G X, WANG X S, CHENG Y H. Hyperspectral image classification based on local binary pattern and broad lear-ning system[J]. International Journal of Remote Sensing, 2020, 41(24): 9393-9417. DOI URL
[5]	IQBAL M, ASIFULLAH K, NAEEM A. Object detection using hybridization of static and dynamic feature spaces and its exploitation by ensemble classification[J]. Neural Compu-ting and Applications, 2019, 31(2): 347-361.
[6]	KUSHBU S C, INBAMALAR T M. Interactive one way con-tour initialization for cardiac left ventricle and right vent-ricle segmentation using hybrid method[J]. Journal of Medical Imaging and Health Informatics, 2021, 11(4): 1037-1054.
[7]	ZHAO Y, OUYANG P, KANG W, et al. An STT-MRAM based in memory architecture for low power integral computing[J]. IEEE Transactions on Computers, 2019, 68(4): 617-623. DOI URL
[8]	YANG T, SHI H Y, LANG M Y, et al. ISAR imaging enhan-cement: exploiting deep convolutional neural network for signal reconstruction[J]. International Journal of Remote Sen-sing, 2020, 41(24): 9447-9468.
[9]	BOWDEN D C, SAGER K, FICHTNER A, et al. Connec-ting beamforming and kernel-based noise source inversion[J]. Geophysical Journal International, 2021, 224(3): 1607-1620. DOI URL
[10]	DHAREJO F A, ZHOU Y C, DEEBA F, et al. A remote-sensing image enhancement algorithm based on patch-wise dark channel prior and histogram equalisation with colour correction[J]. IET Image Processing, 2020, 15(1): 47-56. DOI URL
[11]	VORHIES J T, HOOVER A P, MADANAYAKE A. Adap-tive filtering of 4-D light field images for depth-based image enhancement[J]. IEEE Transactions on Circuits and Systems II-Express Briefs, 2021, 68(2): 787-791. DOI URL
[12]	SOLER J D. Using herschel and planck observations to deli-neate the role of magnetic fields in molecular cloud struc-ture[J]. Astronomy and Astrophysics, 2019, 629(A96): 1-26.
[13]	王化喆, 李德启. 空间域下基于直方图处理的图像增强算法[J]. 计算机光盘软件与应用, 2013(18): 96-97.
	WANG H Z, LI D Q. Image enhancement algorithm based on histogram processing in spatial domain[J]. Computer CD Software and Applications, 2013(18): 96-97.
[14]	LEI H, CHANG X Y, WANG F, et al. A novel algorithm based on histogram processing of reliability for two-dimen-sional phase unwrapping[J]. Optik, 2015, 126(18): 1640-1644. DOI URL
[15]	YANG J Q, ZHANG Q, CAO Z G. Multi-attribute statistics histograms for accurate and robust pairwise registration of range images[J]. Neurocomputing, 2017, 251(4): 54-67. DOI URL
[16]	CHU H, TONG Z, ZHAO S, et al. Particle filter sample texton feature for SAR image classification[J]. IEEE Geo-science and Remote Sensing Letters, 2015, 12(5): 1141-1145.
[17]	王勇, 刘雯. 改进的图像增强直方图统计方法[J]. 吉林大学学报(信息科学版), 2015, 33(5): 495-500.
	WANG Y, LIU W. Improved image enhancement histogram-equalization method[J]. Journal of Jilin University (Information Science Edition), 2015, 33(5): 495-500.
[18]	YE L N, HOU Z J, ENG H L. Context aware image enhance-ment for online fish behaviour monitoring[J]. IET Image Processing, 2016, 10(2): 149-157. DOI URL
[19]	ZHU C, WANG J, LIU H, et al. Insect identification and counting in stored grain: image processing approach and application embedded in smartphones[J]. Mobile Informa-tion Systems, 2018, 24(6): 1-5.
[20]	周启双, 孙玉秋. 基于直方图统计的自适应图像增强改进算法[J]. 长江大学学报(自然科学版), 2015, 12(1): 49-56.
	ZHOU Q S, SUN Y Q. The enhancement algorithm of local contrast based on the statistics of histogram[J]. Journal of Yangtze University (Natural Science Edition), 2015, 12(1): 49-56.
[21]	HE C, YE Y P, TIAN L, et al. A statistical distribution texton feature for synthetic aperture radar image classifica-tion[J]. Frontiers of Information Technology and Electronic Engineering, 2017, 18(10): 1614-1623. DOI URL
[22]	吕新正, 张敏. 高密度环境下的脉冲分选技术研究[J]. 航天电子对抗, 2020, 36(1): 50-53.
	LYU X Z, ZHANG M. The technology research of pulse sorting based on high density environment[J]. Aerospace Elec-tronic Warfare, 2020, 36(1): 50-53.
[23]	YANG R H, LI Q T, TAN J X, et al. Accurate road marking detection from noisy point clouds acquired by low-cost mobile LiDAR systems[J]. ISPRS International Journal of Geo-Information, 2020, 9(10): 1-14. DOI URL
[24]	胡英帅. Shape Context算法的CUDA并行化实现[J]. 电脑编程技巧与维护, 2017(14): 5-7.
	HU Y S. CUDA parallel implementation of Shape Context algorithm[J]. Computer Programming Skills and Maintenance, 2017(14): 5-7.
[25]	KARBOWIAK Ł. Improving efficiency of automatic labe-ling by image transformations on CPU and GPU[C]// LNCS 12043: Proceedings of the 13th International Conference on Parallel Processing and Applied Mathematics, Poland, Sep 8-11, 2019. Cham: Springer, 2019: 479-490.
[26]	GOCHO M, ARII M. Efficient GPU-based local histogram analyzer for change detection in satellite SAR images[C]// Proceedings of the 2020 IEEE International Geoscience and Remote Sensing Symposium, Waikoloa, Sep 26-Oct 2, 2020. Piscataway: IEEE, 2020: 304-307.
[27]	裴浩, 游小荣, 牛欣伟. 矿山三维空间数据距离直方图算法优化及加速[J]. 工矿自动化, 2017, 43(2): 55-60.
	PEI H, YOU X R, NIU X W. Optimization and acceleration of distance histogram algorithm of three-dimension space data of coal mine[J]. Industry and Mine Automation, 2017, 43(2): 55-60.
[28]	陈坤, 黎煊, 刘品, 等. 基于FPGA 的图像形状边缘匹配算法的实现[J]. 科技创新与应用, 2016(11): 1-2.
	CHEN K, LI X, LIU P, et al. Implementation of image shape edge matching algorithm based on FPGA[J]. Techno-logy Innovation and Application, 2016(11): 1-2.
[29]	HONZATKO D, KRULIS M. Accelerating block-matching and 3D filtering method for image denoising on GPUs[J]. Jou-rnal of Real-Time Image Processing, 2019, 16(6): 2273-2287.
[30]	NAYAGAM M G, RAMAR K. Reliable object recognition system for cloud video data based on LDP features[J]. Com-puter Communications, 2020, 149(1): 343-349.
[31]	CHEN A G, YAN H Y. An improved fuzzy C-means cluste-ring for brain MR images segmentation[J]. Journal of Medi-cal Imaging and Health Informatics, 2021, 11(2): 386-390.
[32]	RAD H S, SAFARI M, KAZEROONI A F, et al. Apparent diffusion coefficient (ADC) and first-order histogram statis-tics in differentiating malignant versus benign meningioma in adults[J]. Iranian Journal of Radiology, 2019, 16(1): 1-9.
[33]	JIANG N, JI Z X, WANG J, et al. Quantum image histo-gram statistics[J]. International Journal of Theoretical Physics, 2020, 59(11): 3533-3548. DOI URL

面向GPU的直方图统计图像增强并行算法

GPU-Oriented Parallel Algorithm for Histogram Statistical Image Enhancement

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 33

相关文章 12

编辑推荐

Metrics

[1]	文敏华，刘永志，鲍华，胡跃，沈泳星，韦建文，林新华. 声子BTE应用的并行和优化研究[J]. 计算机科学与探索, 2020, 14(8): 1288-1297.
[2]	武海鹏，文敏华，SEE Simon，林新华. 激光等离子体相互作用模拟的并行和加速研究[J]. 计算机科学与探索, 2018, 12(4): 550-558.
[3]	陈思汉，余建波. 基于二维局部均值分解的图像边缘检测算法[J]. 计算机科学与探索, 2016, 10(6): 847-855.
[4]	王虹旭，吴斌，刘旸. 基于Spark的并行图数据分析系统[J]. 计算机科学与探索, 2015, 9(9): 1066-1074.
[5]	陈军. 微波源器件模拟中的并行FDTD建模[J]. 计算机科学与探索, 2015, 9(11): 1295-1300.
[6]	贾伟乐，曹宗雁，王龙，迟学斌，高卫国，汪林望. Ultra-Mat：基于平面波的第一原理异构计算软件[J]. 计算机科学与探索, 2014, 8(7): 769-777.
[7]	韦向远，杨辉华，谢谱模. 基于CUDA的并行布谷鸟搜索算法设计与实现[J]. 计算机科学与探索, 2014, 8(6): 665-673.
[8]	覃子姗，顾璠，秦晓科，陈铭松. 基于GPU平台的有效字典压缩与解压缩技术[J]. 计算机科学与探索, 2014, 8(5): 525-536.
[9]	覃雄派. 星型模型上的高效百分点计算方法[J]. 计算机科学与探索, 2013, 7(5): 385-393.
[10]	文敏华，林新华，Simon Chong Wee See. 动态网格的DSMC方法在GPU上的并行[J]. 计算机科学与探索, 2013, 7(5): 472-479.
[11]	周国亮，王桂兰，朱永利. 多核处理器上的并行联机分析处理算法研究[J]. 计算机科学与探索, 2013, 7(2): 180-190.
[12]	董辉+,马垣,宫玺. 一种新的概念格并行构造算法[J]. 计算机科学与探索, 2008, 2(6): 651-657.