臺灣博碩士論文加值系統

科技的發展以及硬體上的進步，影像被廣泛用於3D影像和加速遊戲應用程序，圖形處理器Graphics Processing Unit(GPU)專為密集且並行化的設計進行計算，隨著統一計算設備架構Compute Unified Device Architecture(CUDA)的整合技術崛起，使得GPU高性能應用的功能被展現出來。過往的GPU主要負責圖像的處理，現在隨著人工智慧的影響下，開發需求上的準確率與速度已經遠遠超過以往的需求，運算的調節慢慢趨向飽和狀態，目前能夠再提高效能的方式就是運算單元平行化。

影像處理在科技業上的應用也越來越多，而本研究在科技業中實作CUDA、Python以及OpenCV三者的影像處理效能，探索GPU加速在影像處理上與CPU之間的效能，進而了解到GPU加速的差距。

With the development of technology and hardware advancement, images are widely used in 3D imaging and accelerated game applications. GPU is designed for intensive and parallel design for computing. With the rise of CUDA unified computing device architecture integration technology, GPU is the function of the performance application is revealed.

In the past, GPU was mainly responsible for image processing. Now, under the influence of artificial intelligence, the accuracy and speed of development requirements have far exceeded the previous requirements. The adjustment of calculations is gradually becoming saturated, and it is currently improving performance. The way is to parallelize the computing unit.

There are more and more applications of image processing in the technology industry, and this research has implemented the image processing performance of CUDA, Python and OpenCV in the technology industry, exploring the performance of GPU acceleration in image processing and CPU, and then Understand the gap of GPU acceleration.

中文摘要 ...i
英文摘要 ...ii
誌謝 ...iii
目錄 ...iv
表目錄 ...vi
圖目錄 ...vii
第一章 緒論...1
1.1 研究背景 ...1
1.2 研究動機與目的...2
1.3 研究流程 ...2
第二章 文獻探討 ...4
2.1 GPU...4
2.2 CUDA...5
2.2.1 HOST與DEVICE...5
2.2.2 組織架構 ...6
2.2.3 Memory Hierarchy...6
2.3 數位影像處理...8
2.3.1 影像卷積(Convolution)...8
2.3.2 索伯運算(Sobel Operator) ...8
2.3.3 高斯模糊(Gaussian Blur) ...9
2.4 RAW DATA...11
第三章 研究方法與實作...12
3.1 研究架構 ...12
3.2 影像預處理...13
3.2.1 影像轉換 ...13
3.2.2 影像驗證 ...15
3.3 CUDA ...15
3.3.1 HOST與DEVICE...15
3.3.2 CUDA函式...16
3.3.3 線程組織 ...17
3.4 影像處理 ...17
3.4.1 影像卷積 ...17
3.4.2 索伯運算 ...18
3.4.3 高斯模糊 ...19
第四章 研究分析與結果...21
4.1 運算環境 ...21
4.1.1 Windows ...21
4.1.2 AGX Xavier...21
4.2 運行分析 ...22
4.2.1 影像卷積效能分析...22
4.2.2 索伯演算法效能分析...27
4.2.3 高斯模糊演算法效能分析...32
4.3 研究比較 ...37
第五章 結論與未來展望...39
5.1 結論...39
5.2 未來展望 ...39
參考文獻 ...40
Extended Abstract...42

[1]NVIDIA CUDA C Programming Guide, Version 11.0
https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html

[2]誠君. (2005) CPU的下一步
https://www.ctimes.com.tw/DispCols/tw/05010318395S.shtml

[3]Owens, J. D., Houston, M., Luebke, D., Green, S., Stone, J. E., & Phillips, J. C. (2008). GPU computing. Proceedings of the IEEE, 96(5), 879-899.

[4]Ramamoorthy, C. V., & Li, H. F. (1977). Pipeline architecture. ACM Computing Surveys (CSUR), 9(1), 61-102.

[5]Yang, Z., Zhu, Y., & Pu, Y. (2008, December). Parallel image processing based on CUDA. In 2008 International Conference on Computer Science and Software Engineering 198-201.

[6]李堅萍. (2005). 應用數位影像處理科技之影像模擬功能於視覺藝術教學與設計實務. 生活科技教育, 38(5), 26-31.

[7]LeCun, Y., & Bengio, Y. (1995). Convolutional networks for images, speech, and time series. The handbook of brain theory and neural networks, 3361(10).

[8]Sobel, I., & Feldman, G. (1968). A 3x3 isotropic gradient operator for image processing,” a talk Stanford Artif.

[9]Hsiao, P. Y., Chou, S. S., & Huang, F. C. (2007). Generic 2-d gaussian smoothing filter for noisy image processing. In TENCON 2007-2007 IEEE Region 10 Conference.

[10]Liang, C. H., Chen, Y. A., Liu, Y. C., & Hsu, W. H. (2020). Raw Image Deblurring. arXiv preprint arXiv:2012.04264.

[11]NVIDIA Jetson AGX Xavier 開發套件與模組
https://www.nvidia.com/zh-tw/autonomous-machines/embedded-systems/jetson-agx-xavier/

[12]Tse, J. J. (2012). Image processing with CUDA.

[13]Podlozhnyuk, V. (2007). Image convolution with CUDA. NVIDIA Corporation white paper, June, 2097(3).

[14]Russo, L. M., Pedrino, E. C., Kato, E., & Roda, V. O. (2012, March). Image convolution processing: A GPU versus FPGA comparison. In 2012 VIII southern conference on programmable logic 1-6.