因為交錯影像格式能夠有效減少傳輸影像的頻寬使用量，其技術現今仍然使用
在許多廣播系統上(如NTSC、PAL、SECOM)。然而現今已廣泛使用的逐行掃描
顯示裝置，並無法支援這種格式的影像，因此，交錯影像的重建一直是一個重要
待解決的課題。在本篇論文中，我們提出基於深度學習與運動補償之影像去交錯
演算法。我們的方法是採用兩階段式填值達到影像去交錯。首先，對於輸入的交
錯影像使用自我驗證法(Self Validation) 來做初步影像去交錯得到第1 階段的填
值；接著使用運動適應法區分初步影像去交錯得到的像素是位於靜態區塊或動態
區塊。對於動態區塊的像素，我們會直接採用初步影像去交錯的結果，對於靜態
區塊的像素則會經過模組化類神經網路(Modular Neural Networks) 做第2 階段的
影像去交錯改善填值效果。實驗的部分，我們使用10 部CIF 影像與其他演算法
做客觀性能評比，實驗結果證明我們提出的演算法優於其他演算法結果。

Since interlaced video format can effectively reduce the usage of bandwidth in video transmission, this technical is widely used on many Broadcast systems, such as NTSC, PAL, and SECOM. However, progressive scan devices in general do not support this format and need to combine two fields of interlaced video into a single frame, which leads to visual defects. Therefore high quality video deinterlacing method that minimizes these defects is in need.

In this thesis, we propose a novel video deinterlacing algorithm base on deep learning and motion compensation. Our proposed video deinterlacing method consists of two steps: first, self validation is employed to determine the initial deinterlaced result from various candidates; second, the static region of the initial deinterlaced result is identified and further refined based on the modular neural networks.

In the experimental section, 10 popular CIF video sequences are utilized for the objective performance evaluation. The experimental results show that our method outperforms the other deinterlacing methods.

[1] G. de Haan and E. B. Bellers, “Deinterlacing-an overview,” Proc. IEEE, vol. 86, p. 1839–1857, 1998.
[2] K. N. M. K. Park, M. G. Kang and S. G. Oh, “New edge dependent deinterlacing algorithm based on horizontal edge pattern,” IEEE Transactions
on Consumer Electron, vol. 49, p. 1508–1512, 2003.
[3] S. H. L. H. I. Koo and N. I. Cho, “A new edi-based deinterlacing algorithm,” IEEE Transactions on Consumer Electron, vol. 53, p. 1494–1499, 2007.
[4] J. J. Jin Wang, Gwanggil Jeon, “Moving least-squares method for interlaced to progressive scanning format conversion,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 23, pp. 1865 – 1872, 2013.
[5] Y. H. L. S. Yang, Y. Y. Jung and R. H. Park, “Motion compensation assisted motion adaptive interlaced-to-progressive conversion,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 14, pp. 1138–1148, 2004.
[6] K. L. C. Lee, “de-interlacing with motion adaptive vertical temporal filtering,” IEEE Transactions on Consumer Electron, vol. 55, p. 636–643, 2009.
[7] H. Choi and C. Lee, “Motion adaptive de-interlacing with modular neural networks,” IEEE Transactions on Circuits and Systems for Video Technology,
vol. 21, p. 844–849, 2011.
[8] P. L. H. M. Mohammadi and Y. Savaria, “A five-field motion compensated deinterlacing method based on vertical motion,” IEEE Transactions on Consumer Electron, vol. 53, p. 1117–1124, 2007.
[9] Y.-C. Fan and C.-H. Chung, “De-interlacing algorithm using spatial-temporal correlation-assisted motion estimation,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 19, p. 932–944, 2009.
[10] S. M. Q. Huang, D. Zhao and H. Sun, “de-interlacing using hierarchical motion analysis,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 20, p. 673–686, 2010.
[11] S.-Y. C. Ting-Chun Wang, Yi-Nung Liu, “Algorithm adaptive video deinterlacing using self-validation framework,” IEEE International Symposium on Circuits and Systems, p. 2804–2807, 2013.
[12] K. L. C. Lee, “High quality spatially registered vertical temporal filtering for deinterlacing,” IEEE Transactions on Consumer Electronics, vol. 59, pp. 182–190, 2013.
[13] R. Dehghannasiri, “Video de-interlacing using asymmetric nonlocal-means filtering,” 48th Asilomar Conference on Signals, Systems and Computers, p. 688–692, 2014.
[14] S. S. Farhang Vedadi, “A new map-based approach to video de-interlacing using forward-backward algorithm,” 46th Asilomar Conference on Signals, Systems and Computers, p. 1703–1707, 2012.
[15] R. Dehghannasiri and S. Shirani, “A novel de-interlacing method based on locally-adaptive nonlocal-means,” 46th Asilomar Conference on Signals, Systems and Computers, pp. 1708–1712, 2012.