臺灣博碩士論文加值系統

影像降噪（image denoising）是影像處理（image processing）領域中非常重要之研究議題。近年隨著智慧型手機、網際網路、以及社群媒體的蓬勃發展與普及化，透過數位影像傳遞資訊之方式逐漸廣為使用，而影像降噪相關等可提升影像品質之演算法顯得更加重要。

本論文中，延伸自 Ulyanov 等人~\cite{Ulyanov_2018_CVPR} 對於卷積神經網路（convolutional neural networks）架構性質之研究，進行卷積神經網路架構性質與降噪影像生成關係之實驗，對於卷積神經網路架構中各項運算子給出簡單且詳盡之概述，並針對此方法設計模型架構，且採用特殊正規化方式提升影像降噪之效率。

其中，此方法與近年主流卷積神經網路於影像應用之研究較為不同，主要的差異在於不需透過大量且成對之資料集進行訓練，訓練時間相對較短，且模型僅向具噪音之受損影像（noisy image）進行學習與生成，自始至終模型並未看過原始品質良好之影像（ground truth）。

Removing noise from the images to improve image quality is the main challenge in image processing. Especially as the ubiquitous spread of computers, smartphones, the Internet, and social networks, image denoising becomes more and more important.

In this work, we extend upon the results of Ulyanov et al.~\cite{Ulyanov_2018_CVPR} and introduce a competitive image denoising method based on the structure characteristic of convolutional neural networks (CNNs). Different from most CNN-based methods which need a large-scale dataset for training, our method only looks at one degraded image and removes noise on itself. This method is not only an application of image denoising but also a point of view for visualizing the property and effect of each element in convolutional neural networks.

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Image Reconstruction Characteristic of CNN . . . . . . . . . . . . . . . . 5
2.2 Relevant Image Denoising Methods . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Model-based Methods . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Discriminative Learning-based Methods . . . . . . . . . . . . . . . 7
3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1 Image Denoising Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Image Denoising Process . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4 Model Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.1 Basic Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.1.1 Convolutional Layer . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.1.2 Padding Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.1.3 Normalization Layer . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.1.4 Activation Function . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 High-level Blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2.1 Downsampling Block . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.2 Skip-connect Block . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2.3 Upsampling Block . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5 Experiment and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

References
[Buad2005] A. Buades, B. Coll, and J.-M. Morel. A non-local algorithm for image denoising.
IEEE Conference on Computer Vision and Pattern Recognition
(CVPR), volume 2, 60–65, 2005.
[Dabo2007] K. Dabov, A. Foi, V. Katkovnik, and K. Egiazarian. Image denoising by
sparse 3-d transform-domain collaborative filtering. IEEE Transactions on
Image Processing, 16(8):2080–2095, 2007.
[Glor2010] X. Glorot and Y. Bengio. Understanding the difficulty of training deep feedforward
neural networks. Proceedings of the Thirteenth International Conference
on Artificial Intelligence and Statistics, volume 9 of Proceedings of
Machine Learning Research, 249–256, 2010.
[Good2014] I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S.
Ozair, A. Courville, and Y. Bengio. Generative adversarial nets. Advances
in Neural Information Processing Systems 27 (NIPS), 2672–2680, 2014.
[He2016] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image
recognition. IEEE Conference on Computer Vision and Pattern Recognition
(CVPR), 770–778, 2016.
[Ioff2015] S. Ioffe and C. Szegedy. Batch normalization: accelerating deep network
training by reducing internal covariate shift. Proceedings of the 32nd International
Conference on Machine Learning (ICML), volume 37 of Proceedings
of Machine Learning Research, 448–456, 2015.
[LeCu1989] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard,
and L. D. Jackel. Backpropagation applied to handwritten zip code
recognition. Neural Computation, 1(4):541–551, 1989.
[Lin2014] M. Lin, Q. Chen, and S. Yan. Network in network. Proceedings of the International
Conference on Learning Representations (ICLR), 2014.
[Maas2013] A. L. Maas, A. Y. Hannun, and A. Y. Ng. Rectifier nonlinearities improve
neural network acoustic models. ICML Workshop on Deep Learning for
Audio, Speech, and Language Processing (WDLASL), volume 30, page 3,
2013.
[Rama2018] P. Ramachandran, B. Zoph, and Q. V. Le. Searching for activation functions,
2018. URL: https://openreview.net/forum?id=SkBYYyZRZ.
[Tian2018] C. Tian, Y. Xu, L. Fei, and K. Yan. Deep learning for image denoising: A
survey. CoRR, abs/1810.05052, 2018. URL: http://arxiv.org/abs/
1810.05052.
[Ulya2018] D. Ulyanov, A. Vedaldi, and V. Lempitsky. Deep image prior. IEEE Conference
on Computer Vision and Pattern Recognition (CVPR), 9446–9454,
2018.
[Ulya2016] D. Ulyanov, A. Vedaldi, and V. S. Lempitsky. Instance normalization: the
missing ingredient for fast stylization. CoRR, abs/1607.08022, 2016. URL:
http://arxiv.org/abs/1607.08022.
[Wang2003] Z. Wang, E. P. Simoncelli, and A. C. Bovik. Multiscale structural similarity
for image quality assessment, volume 2, 1398–1402, 2003.
[Wang2004] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli. Image quality
assessment: from error visibility to structural similarity. IEEE Transactions
on Image Processing, 13(4):600–612, 2004.
[Wu2018] Y.Wu and K. He. Group normalization. European Conference on Computer
Vision (ECCV), 3–19, 2018.
[Zhan2017] K. Zhang, W. Zuo, Y. Chen, D. Meng, and L. Zhang. Beyond a gaussian
denoiser: residual learning of deep CNN for image denoising. IEEE Transactions
on Image Processing, 26(7):3142–3155, 2017.
[Zhan2018] K. Zhang,W. Zuo, and L. Zhang. FFDNet: toward a fast and flexible solution
for CNN based image denoising. IEEE Transactions on Image Processing,
27(9):4608–4622, 2018.
[Abad2015] M. Abadi et al. TensorFlow: large-scale machine learning on heterogeneous
systems, 2015. URL: https://www.tensorflow.org/. Software
available from tensorflow.org.
[Web01] M. Colom. NLM image denoising demo. http://demo.ipol.im/
demo/bcm_non_local_means_denoising. last accessed: 2018-12-
31.
[Web02] M. Lebrun. BM3D image denoising code. https://github.com/
gfacciol/bm3d. last accessed: 2018-12-31.