研究生: |
林立東 Li-Tung-Lin |
---|---|
論文名稱: |
基於邊緣導向內插超解析成像方法應用於行人影像 Super-Resolution Imaging Method Based on Edge-Directed Interpolation Applied to Pedestrian Images |
指導教授: |
王乃堅
Nai-Jian Wang |
口試委員: |
郭景明
Jing-Ming Guo 蘇順豐 Shun-Feng Su |
學位類別: |
碩士 Master |
系所名稱: |
電資學院 - 電機工程系 Department of Electrical Engineering |
論文出版年: | 2023 |
畢業學年度: | 111 |
語文別: | 中文 |
論文頁數: | 59 |
中文關鍵詞: | 超解析成像 、邊緣導向插值法 、梯度直方圖 、支持向量機 |
外文關鍵詞: | Super-Resolution, NEDI, HOG, SVM |
相關次數: | 點閱:217 下載:5 |
分享至: |
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
現今的科技日新月異,視訊監視系統的普及使得各種視訊監控影像被廣泛使用,然而路口監視器或是其他影像感測器常因為低品質的監視影像而無法有良好的應用效果,因此本論文將針對視訊監控影像的行人部分進行超解析度成像。
為了使得視訊監控影像能夠獲得更多的影像細節,提高解析度有利於後續電腦視覺的影像分析與應用,因此本系統將著重於超解析度成像的品質,並利用超解析度成像將行人影像中行人的部分提高其解析度。
本論文完成了一個針對視訊監控影像中行人部分的超解析度方法,首先會以梯度直方圖 (Histograms of Oriented Gradients)和支持向量機(Support Vector Machine)的方式進行視訊監控影像中行人的檢測,並採用修改後的邊緣導向插值法MNEDI (Modify New Edge Directed Interpolation) 提高行人影像中行人部分的解析度。本系統首先利用邊緣偵測,將影像切割為平滑區域與邊緣區域,分別做不同的插值計算,相較傳統演算法中能夠抑制超解析度成像後造成邊緣的鋸齒狀情形,以及平滑區域中的詭影情形,以達到高品質的超解析度效果。
我們在包含大量複雜環境下的行人影像測試數據集上,還有最常用來比對成像品質的數據集Set-5進行了實驗驗證,實驗結果顯示,相較於現有的方法,本研究提出的演算法在非神經網路的超解析成像的影像放大質量方面表現優異,適合應用於智能監控和車輛主動安全等電腦視覺領域。
In today's rapidly advancing technology, the widespread of video surveillance systems has led to extensive utilization of various video monitoring imagery. However, surveillance cameras at intersections or other image sensors often suffer from low-quality surveillance video, resulting in suboptimal application effectiveness. Therefore, this thesis focus on address super-resolution imaging for the pedestrian parts of video surveillance imagery.
To capture more image details and enhance the resolution for subsequent computer vision analysis and applications, this system emphasizes the quality of super-resolution imaging. By utilizing super-resolution techniques, the resolution of pedestrian parts within the surveillance imagery can be enhanced.
This thesis presents a super-resolution system specifically tailored for the pedestrian parts of video surveillance imagery. First, pedestrian detection is performed using Histograms of Oriented Gradients (HOG) and Support Vector Machine (SVM) techniques. Subsequently, a modified version of the New Edge Directed Interpolation (MNEDI) algorithm is employed to enhance the resolution of the pedestrian parts. The system employs edge detection to segment the image into smooth and edge regions, applying different interpolation calculations to each region. Compared to traditional algorithms, this method effectively suppresses artifacts such as jagged edges along the edges and ghosting in smooth regions, produce high-quality image.
Experimental validation is conducted on a large dataset containing pedestrian images in complex environments, including the widely used Set-5 dataset for image quality comparison. The experimental results demonstrate the superior performance of the proposed algorithm in non-neural-network-based super-resolution image enlargement quality. This makes it well-suited for applications in computer vision domains such as intelligent surveillance and vehicle active safety.
[1] X. Li and M. Orchard ,"New Edge Directed Interpolation," Proceedings 2000 International Conference on Image Processing ,(2000), pp. 311-314
[2] N. Dalal and B. Triggs, "Histograms of Oriented Gradients for Human Detection," 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), (2005), pp. 886-893
[3] Y. Luo, S. Liu and H. Zhu, "Edge-Directed Interpolation based On Canny Detector," 2011 IEEE International Conference on Mechatronics and Automation, (2011), pp. 698-702
[4] M. Tipping and C. Bishop, “Bayesian Image Super-Resolution,” in Advances in Neural Information Processing Systems, (2003), pp. 1303– 1310.
[5] L. Zhang, H. Zhang, H. Shen, and P. Li, “A Super-Resolution Reconstruction Algorithm for Surveillance Images,” Signal Processing, (2010),vol. 90, no. 3, pp. 848–859,
[6] M. S. Alam, J. G. Bognar, R. C. Hardie, and B. J. Yasuda, “Infrared Image Registration and High-Resolution Reconstruction using Multiple Translationally Shifted Aliased Video Frames,” IEEE Transactions on Instrumentation and Measurement, (2000) vol. 49, no. 5, pp. 915–923,
[7] C. Dong, C. C. Loy, K. He, and X. Tang, “Learning a Deep Convolutional Network for Image Super-Resolution,” in European Conference on Computer Vision.
Springer, (2014), pp. 184–199.
[8] C. Dong, C. C. Loy, and X. Tang, “Accelerating the Super-Resolution Convolutional Neural Network,” in European Conference on Computer Vision. Springer, (2016), pp. 391–407.
[9] Ledig, Christian, et al. "Photo-Realistic Single Image Super-Resolution using a Generative Adversarial Network." Proceedings of the IEEE conference on computer vision and pattern recognition. (2017).
[10] G. Jing, Y. Choi, K. Wang and W. Wang. " Gradient Guided Image Interpolation “2014 IEEE International Conference on Image Processing (ICIP) (2014). pp. 1822-1826
[11] R. Keys. "Cubic Convolution Interpolation for Digital Image Processing." IEEE transactions on acoustics, speech, and signal processing . (1981) , pp. 1153-1160
[12] Smith, P. R. "Bilinear Interpolation of Digital Images." Ultramicroscopy 6.2 (1981): 201-204.
[13] S. Wee, D. Park and J. Jeong, " Modified New Edge-Directed Interpolation Using Window Extension." 2017 International Journal of Mathematical and Computational Methods, (2017).
[14] N. Asuni and A. Giachetti, "Accuracy Improvements and Artifacts Removal In Edge Based Image Interpolation,“2008 Computer Vision Theory and Applications, Madeira, Jan. (2008).