计算机科学 ›› 2014, Vol. 41 ›› Issue (5): 14-19.doi: 10.11896/j.issn.1002-137X.2014.05.003

• 综述 • 上一篇    下一篇

CPU-GPU协同计算加速ASIFT算法

何婷婷,芮建武,温腊   

  1. 中国科学院软件研究所 北京100190;中国科学院软件研究所 北京100190;中国科学院软件研究所 北京100190
  • 出版日期:2018-11-14 发布日期:2018-11-14
  • 基金资助:
    本文受云计算操作系统及关键基础组件的研究与开发:面向云计算的大数据集并行处理平台研究与开发(KGCX2-YW-174),国家科技支撑计划项目:新型网络终端操作系统社区版本研究与开发,应用程序库汇总meegobox(2011BAH14B02),2012年度“核高基”重大专项:开源操作系统内核分析和安全性评估(2012ZX01039-002),新闻出版重大科技工程项目-中华字库工程-第23包:应用平台研发(GAPP-ZDKJ-ZK/23)资助

Accelerating ASIFT Based on CPU/GPU Synergetic Parallel Computing

HE Ting-ting,RUI Jian-wu and WEN La   

  • Online:2018-11-14 Published:2018-11-14

摘要: ASIFT(Affine-SIFT)是一种具有仿射不变性、尺度不变性的特征提取算法,其被用于图像匹配中,具有较好的匹配效果,但因计算复杂度高而难以运用到实时处理中。在分析ASIFT算法运行耗时分布的基础上,先对SIFT算法进行了GPU优化,通过使用共享内存、合并访存,提高了数据访问效率。之后对ASIFT计算中的其它部分进行GPU优化,形成GASIFT。整个GASIFT计算过程中使用显存池来减少对显存的申请和释放。最后分别在CPU/GPU协同工作的两种方式上进行了尝试。实验表明,CPU负责逻辑计算、GPU负责并行计算的模式最适合于GASIFT计算,在该模式下GASIFT有很好的加速效果,尤其针对大、中图片。对于2048*1536的大图片,GASIFT与标准ASIFT相比 加速比 可达16倍,与OpenMP优化过的ASIFT相比加速比 可达7倍,极大地提高了ASIFT在实时计算中应用的可能性。

关键词: 特征提取,ASIFT,SIFT,CPU/GPU协同工作

Abstract: ASIFT(affine-SIFT) is a fully affine invariant,and scale invariant image local feature extraction algorithm.It has a good result in image matching.But because of its high computational complexity,it cannot be applied to real-time processing.Thus GPU is used to accelerate ASIFT.Based on the analysis of running time of ASIFT,firstly SIFT was adapted to GPU,and then the other parts of ASIFT.Memory pool was used in GASIFT to avoid frequently allocating and deleting memory during the runtime.Different ways of CPU/GPU synergetic parallel computing were studied to make GASIFT more efficient.Experiments show that the model in which CPU takes the logical calculation work and GPU makes parallel computing is the most suitable way.Based on this model,GASIFT has a good speed-up ratio over other methods.That’s 16times compared with traditional ASIFT,and 7times compared with OpenMP optimized ASIFT.

Key words: Image feature extraction,ASIFT,SIFT,CPU/GPU synergetic parallel

[1] Lowe D G.Distinctive image features from scale-invariant keypoints [J].International Journal of Computer Vision,2004,60(2):91-110
[2] Yan Ke.PCA-SIFT:A more distinctive representation for local image descriptors[C]∥ CVPR .Washington.DC,USA,2004:66-75
[3] Bay H,Tuytelaars T,Van Gool L.SURF:Speeded up robustfeatures,2006[C]∥Proc.European Conference on Computer Vision.2006:404-417
[4] Heymann S,Maller K,Smolic A,et al.SIFT implementation and optimization for general-purpose GPU,2007[C]∥Proc.International Conference in Central Europe on Computer Graphics,Visualization and Computer Vision.2007:1-8
[5] 王瑞,梁华,蔡宣平,基于GPU的SIFT特征提取算法研究 [J].现代电子技术,2010,33(15):41-46
[6] Daason K,Lejsek H,rsll T,et al.GPU acceleration of Eff2descriptors using CUDA,2010[C]∥ Proceedings of the International Conference on Multimedia.Firenze,Italy,2010:1167-1170
[7] Morel J M,Yu G.ASIFT:A New Framework for Fully Affine Invariant Image Comparison [J].SIAM Journal on Imaging Sciences,2009,2(2):1597-1600
[8] Hare J S,Samangooei S,Dupplaw D P.OpenIMAJ and ImageTerrier:Java libraries and tools for scalable multimedia analysis and indexing of images,2011[C]∥ Proceedings of the 19th ACM international conference on Multimedia.Scottsdale,Arizona,USA,2011:691-694
[9] Chu Bin,Jiang Da-lin.Panoramic Image Stitching Using ASIFT,2012[C]∥Fourth International Conference on Multimedia Information Networking and Security.2010:216-219
[10] Yin Chun-xia,Li Cheng-rong,Liu Hong-lin,et al.Experimental Contrast of Several Typical Algorithms for Local Features Detection,2012[C]∥International Conference on Mechanical Engineering and Automation Advances in Biomedical Engineering.2012:65-71
[11] Panga Yan-wei,Lia Wei,Yuan Yuan,et al.Fully affine invariant SURF for image matching[J].Neurocomputing,2012,85:6-10
[12] 卢风顺,宋君强,银福康,等.CPU/GPU协同并行计算研究综述[J].计算机科学,2011,8(3):5-9
[13] 王永明,王贵锦.图像局部不变性特征与描述[M].北京:国防工业出版社,2010:79-87
[14] Morel J-M,Yu Guo-shen.ASIFT:online demo.http://www.cmap.polytechnique.fr/~yu/research/ASIFT/demo.html

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!