隨著第五代行動網路的普及與處理器速度的提高，多媒體通信的品質已從基本的 H.264 支援 2K 視訊編碼，提升到 H265/HEVC 支援 4K 畫質，以及最新的 H.266/VVC 支援 8K 畫質視訊通信編碼。其主要應用的技術為區塊轉換編碼 (block transform coding) 以及動量預測 (motion estimation) 分別處理空間域 (spatial redundancy) 以及時間域 (temporal redundancy) 上的冗餘與編碼。近年來深度學習已應用到視訊編碼，稱為深度視訊編碼 (Deep Video Compression, DVC)。DVC 搭配卷積神經網路自動擷取特徵取代傳統視訊編碼的離散餘旋轉換 (Discrete Cosine Transform, DCT)，可以達到更高的視訊壓縮率。然而卷積神經網路具有特徵冗餘以及梯度爆炸等潛在缺點，考量 DCT 具備正交性，因此本論文探討正交卷積神經網路 OCNN 對編碼效率的影響，於模型訓練時加上卷積層正交規範 (orthogonal constraint)，可以控制卷積層正交的嚴格程度，來達到優化卷積神經網路的效果。本論文提出的深度視訊壓縮附加正交卷積規範，結合深度視訊壓縮 DVC 與正交卷積神經網路 OCNN 的優勢，運用正交規範提升 DVC 模型裡卷積層提取特徵的效果，從而使整體視訊編碼系統達到更好的品質。實驗結果顯示，與未附加正交卷積規範的 OpenDVC 相比，BDBR 為 −5.536%，BD-PSNR 為 0.158dB。

Multimedia communication quality has been improved with the advance of communication technology, e.g., 5G, and processing power. The H.264 supports 2K video coding and the H.265/HEVC and H.266/VVC support 4K and 8K video compression, respectively. Block transform coding and motion estimation, which are the typical video coding methods, are used to remove spatial redundancy and temporal redundancy respectively. Video coding methods have been further improved by deep learning technologies. In contrast to the traditional Discrete Cosine Transform (DCT), a Deep Video Compression (DVC) framework utilizes convolutional neural networks (CNN) to achieve efficient compression. Though the CNN can automatically find out the best features/transformations, it suffers feature redundancy and gradient explosion problems. In addition, the transformations do not guarantee orthogonality like that of the DCT. In this research, we proposed to impose orthogonal constraints on the convolution process to improve the video compression efficiency based on the DVC framework. Experiments showed that a strictly orthogonal constraint is not necessary. The orthogonal constraints should be well-adjusted under different compression ratios to best improve the compression efficiency. As compared with OpenDVC, which is without orthogonal convolution constraint, experimental results show that the BDBR is −5.536% and the BD-PSNR is 0.158dB.

[1] IBM Cloud Education, “Convolutional neural networks.” https://www.ibm.com/
cloud/learn/convolutional-neural-networks, 2020.
[2] A. Wang, “Convolutional neural networks(cnn) #1 kernel, stride, padding.” https://ssur.cc/SkBAeLbPL, 2019.
[3] E. Tang, “Improving autoencoder performance with pretrained rbms.” https://ssur.cc/qTUz6mEfe, 2020.
[4] G. Lu, W. Ouyang, D. Xu, X. Zhang, C. Cai, and Z. Gao, “Dvc: An end-to-end deep
video compression framework,” 2018.
[5] A. Ranjan and M. J. Black, “Optical flow estimation using a spatial pyramid network,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2720–2729, 2017.
[6] J. Wang, Y. Chen, R. Chakraborty, and S. X. Yu, “Orthogonal convolutional neural
networks,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 11505–11515, 2020.
[7] T. Wiegand, G. J. Sullivan, G. Bjontegaard, and A. Luthra, “Overview of the h. 264/avc video coding standard,” IEEE Transactions on circuits and systems for video technology, vol. 13, no. 7, pp. 560–576, 2003.
[8] G. J. Sullivan, J.-R. Ohm, W.-J. Han, and T. Wiegand, “Overview of the high efficiency video coding (hevc) standard,” IEEE Transactions on circuits and systems for video technology, vol. 22, no. 12, pp. 1649–1668, 2012.
[9] R. Yang, L. Van Gool, and R. Timofte, “Opendvc: An open source implementation of the dvc video compression method,” arXiv preprint arXiv:2006.15862, 2020.
[10] K. Yanai, R. Tanno, and K. Okamoto, “Efficient mobile implementation of a cnn-based object recognition system,” in Proceedings of the 24th ACM international conference on Multimedia, pp. 362–366, 2016
[11] F. Heide, W. Heidrich, and G. Wetzstein, “Fast and flexible convolutional sparse coding,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5135–5143, 2015.
[12] F. Bellard, “Bpg image format.” https://bellard.org/bpg/, 2018.
[13] J. Ballé, V. Laparra, and E. P. Simoncelli, “End-to-end optimized image compression,” arXiv preprint arXiv:1611.01704, 2016.
[14] T. Xue, B. Chen, J. Wu, D. Wei, and W. T. Freeman, “Video enhancement with taskoriented flow,” International Journal of Computer Vision (IJCV), vol. 127, no. 8, pp. 1106–1125, 2019.
[15] M. Abadi, P. Barham, J. Chen, Z. Chen, A. Davis, J. Dean, M. Devin, S. Ghemawat,
G. Irving, M. Isard, et al., “{TensorFlow}: a system for {Large-Scale} machine learning,” in 12th USENIX symposium on operating systems design and implementation
(OSDI 16), pp. 265–283, 2016.
[16] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014.
[17] A. Mercat, M. Viitanen, and J. Vanne, “Uvg dataset: 50/120fps 4k sequences for video codec analysis and development,” in Proceedings of the 11th ACM Multimedia Systems Conference, MMSys ’20, (New York, NY, USA), p. 297–302, Association for
Computing Machinery, 2020.