臺灣博碩士論文加值系統

戶外場景影像中含有大量的資訊，其中在戶外影像是由天空的區域與地面區域的結合，而天際線切開兩個區域；所以若是找出天際線，就可以完美的區分出天空地區的資訊與地面地區的資訊。在天際線的運用上，最開始的研究應用NASA的無人探測車在火星上幫助導航辨識方向旋轉的ㄧ個資訊，後來許多研究開始尋找天際線（天空區域），在一些研究中的應用是透過天際線辨識出場景的位置、可以用來辨識山稜線、影像編輯的前處理，例如：行車紀錄器的路面區域資訊、都市的城市線、風景的輪廓等等。
但因為大多數天際線（天空區域）偵測的研究所使用的場景多半為郊區場景，少數的研究才是以市區為場景，其原因是市區所涵蓋的資訊比郊區複雜更多，例如最先面對的挑戰為建築物大樓與天空相似的問題，或者說有不同形狀的建築物會存在建築內縮的問題（倒三角形），這些都更加深了天際線偵測的困難程度。
本系統資料將包含市區與郊區場景。利用區域切割（Region Segmentation），將影像切割成多組區域再將區域邊界切割成線段，透過機器學習分類（Support Vector Machine）的方式，會取得所有邊界的分類。然後以天際線為天空區域的邊界為前提，使用區域填滿（Region filling）取得天際線。
在機器學習特徵擷取的部分，基於特徵擷取來至區域周圍的邊線上的點，本論文使用色彩資訊（RGB Color Patch）以及紋理資訊（Gray-Level Co-Occurrence Matrix）的兩種特徵混用，再使用主成分分析（Principal component analysis）對特徵後處理；最後本論文所提出的特徵以及分類類別，在分類的實驗最後可以有非常好的結果呈現。

The skyline information in an outdoor image scene can provide a lot information such as the horizon and background information for automatic robots. Most existing works find skylines in suburban scenes because the urban scenes are more diverse than suburban scenes. In our paper, we try to find skylines in not only suburban scenes but also urban scenes. We propose a new method to find skylines using region segmentation and boundary feature classification. In the proposed method, we first apply region segmentation methods to obtain region boundaries. Then we use the segmentation result to split the region boundaries to edges. Afterwards, we perform sampling on the segmented region boundaries. Gray-Level Co-Occurrence Matrix (GLCM), color patch descriptors are extracted as the features of sampled pixels on the boundaries. We apply machine learning techniques and use both suburban and urban scenes to perform training. Support vector machine classifiers are used to classify the boundary. The sampled pixels on the boundaries are classified as sky pixel, ground pixel, or skyline pixel. At last, to use the region filling to get the skyline (or sky region). In the experiments, we have tested the performance of the framework using different combinations of region segmentation methods and features. The proposed method exhibits better detection results compared with existing skyline detection methods.

摘要 i
ABSTRACT v
目錄 vii
圖目錄 ix
表目錄 xi
第一章 緒論 1
1.1 研究動機 1
1.2 相關文獻 2
1.3 系統流程論文架構 3
1.3.1 系統流程圖 4
第二章 區域邊界切割 5
2.1 Statistical Region Merging (SRM) 5
2.1 Simple Linear Iterative Clustering (SLIC) 7
2.2 區域邊緣切割 9
第三章 邊緣特徵與分類器 11
3.1 邊緣特徵 11
3.1.1 Gray-Level Co-Occurrence Matrix (GLCM) 12
3.1.2 Color Patch 15
3.1.3 Combine Features 15
3.2 主成分分析Principal component analysis（PCA） 16
3.3 邊緣特徵訓練 18
第四章 天際線選取 22
4.1 天際線尋找方法 22
4.2 天際線尋找問題 23
第五章 實驗結果與討論 26
5.1 資料集 26
5.2 區域切割的參數選用 27
5.3 SVM訓練模式 29
5.4 特徵的比較 30
5.4.1 RGB通道比較 30
5.4.2 GLCM與Color Patch與混合使用比較 31
5.5 方法比較 32
第六章 結論與未來工作 40
6.1 結論 40
6.2 未來工作 41
參考文獻 43

1. Fang, M. Skyline for video-based virtual rail for vehicle navigation. in Intelligent Vehicles Symposium (1993: Tokyo, Japan). Proceedings of the Intelligent Vehicles' 93 Symposium. 1993.
2. Lie, W.-N., et al., A robust dynamic programming algorithm to extract skyline in images for navigation. Pattern recognition letters, 2005. 26(2): p. 221-230.
3. Bazin, J.-C., et al. Dynamic programming and skyline extraction in catadioptric infrared images. in Robotics and Automation, 2009. ICRA'09. IEEE International Conference on. 2009. IEEE.
4. Ahmad, T., et al. A machine learning approach to horizon line detection using local features. in International Symposium on Visual Computing. 2013. Springer.
5. Yazdanpanah, A.P., et al. Sky segmentation by fusing clustering with neural networks. in International Symposium on Visual Computing. 2013. Springer.
6. Hung, Y.-L., et al. Skyline localization for mountain images. in 2013 IEEE International Conference on Multimedia and Expo (ICME). 2013. IEEE.
7. Ahmad, T., et al. An Edge-Less Approach to Horizon Line Detection. in 2015 IEEE 14th International Conference on Machine Learning and Applications (ICMLA). 2015. IEEE.
8. Yazdanpanah, A.P., et al., Real-Time Horizon Line Detection based on Fusion of Classification and Clustering. International Journal of Computer Applications, 2015. 121(10).
9. Shen, Y. and Q. Wang, Sky Region Detection in a Single Image for Autonomous Ground Robot Navigation. International Journal of Advanced Robotic Systems, 2013. 10.
10. Zhijie, Z., et al., A Novel Sky Region Detection Algorithm Based On Border Points. International Journal of Signal Processing, Image Processing and Pattern Recognition, 2015. 8(3): p. 281-290.
11. Nock, R. and F. Nielsen, Statistical region merging. IEEE Transactions on pattern analysis and machine intelligence, 2004. 26(11): p. 1452-1458.
12. Achanta, R., et al., SLIC superpixels compared to state-of-the-art superpixel methods. IEEE transactions on pattern analysis and machine intelligence, 2012. 34(11): p. 2274-2282.
13. Haralick, R.M. and K. Shanmugam, Textural features for image classification. IEEE Transactions on systems, man, and cybernetics, 1973(6): p. 610-621.
14. Long, C.N., et al., Retrieving cloud characteristics from ground-based daytime color all-sky images. Journal of Atmospheric and Oceanic Technology, 2006. 23(5): p. 633-652.
15. Mantelli Neto, S.L., et al., The use of Euclidean geometric distance on RGB color space for the classification of sky and cloud patterns. Journal of Atmospheric and Oceanic Technology, 2010. 27(9): p. 1504-1517.
16. Chang, C.-C. and C.-J. Lin, LIBSVM: a library for support vector machines. ACM Transactions on Intelligent Systems and Technology (TIST), 2011. 2(3): p. 27.
17. Canny, J., A computational approach to edge detection. IEEE Transactions on pattern analysis and machine intelligence, 1986(6): p. 679-698.
18. Jolliffe, I., Principal component analysis. 2002: Wiley Online Library.