一個利用深度卷積網路於動漫影像的超解析技術之研究

簡易檢索 / 詳目顯示

回結果列表

研究生：	陸志良 Chi-Leong Lok
論文名稱：	一個利用深度卷積網路於動漫影像的超解析技術之研究 A Study on the Super Resolution Technique for Animated Images via Deep Convolutional Networks
指導教授：	范欽雄 Chin-Shyurng Fahn
口試委員:	傅楸善 Chiou-Shann Fuh 王聖智 Sheng-Jyh Wang 陳冠宇 Kuan-Yu Chen
學位類別：	碩士 Master
系所名稱：	電資學院 - 資訊工程系 Department of Computer Science and Information Engineering
論文出版年：	2019
畢業學年度：	107
語文別：	英文
論文頁數：	48
中文關鍵詞：	超解析影像放大、動漫影像、深度學習、卷積神經網路、殘差學習、密集連接式卷積神經網路
外文關鍵詞：	Super Resolution, Animated Image, Deep Learning, Convolutional Neural Network, Residual Learning, Dense Convolutional Network
相關次數：	點閱：265 下載：0
分享至:	分享至facebook 分享至twitter

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

最近對超解析度的研究隨著深度卷積神經網絡的發展而取得了進展。儘管使用更快更深的卷積神經網絡在單圖像超解析度的準確性和速度方面取得了突破，但另一方面我們考慮的是如何恢復更精細的紋理細節?然而先前的技術在放大倍數的超解析度時會出現偽影的現像。為了進一步提高視覺品質和優化的超解析度方法，主要藉由目標函數的選擇來達到目的;於本文中，我們開發了一種更深且能重用以前特徵圖的超解析度深度殘差網路，此網路的模型是結合殘差學習 (ResNet) 和密集連接式卷積神經網路 (DenseNet) 兩種架構組合而成，在我們穩定訓練的同時，可以調整訓練影像的大小和影像批次量，它們之間會影響到模型的峰值信號雜訊噪比 (PSNR)，亦在本文中闡述。受益於這些改進，我們的深度殘差網路能夠在公共基準測試裡，從多倍數下採樣的圖像中恢復照片般逼真的紋理，並在公開的數據集方面，我們所提的技術表現優於 SRCNN, VDSR, LapSRN, SRResNet 等方法，根據實驗結果顯示;我們所發展的超解析度深度殘差網路也可以超越 NTIRE2017 的冠軍得主 EDSR 的方法。

Recent research on super resolution has progressed with the development of deep convolutional neural networks (DCNNs). Despite the use of faster and deeper convolutional neural networks, breakthroughs have been made in the accuracy and speed of single image super resolution. On the other hand, how do we restore finer texture details? However, we have artifacts appearing in the large upscale image of super resolution. In order to further improve the visual quality and the optimized super resolution method, the goal is mainly achieved by the selection of the loss function. In this thesis, we have developed a deep residual network that can reuse the previous feature maps. Our purposed model is constructed by the combination of two architectures, ResNet and DenseNet. We can adjust the size of the training image and the batch size during the training. According to this, we can elaborate on them that affect the performance of the deep residual network, say PSNR. Benefiting from these improvements, our deep residual network is able to recover photo-realistic textures from multiple downsampled images in benchmarks, which outperforms the methods such as SRCNN, VDSR, LapSRN, and SRResNet. The experimental results reveal that our developed deep residual network is also better than the NTIRE2017 champion winner EDSR does.

摘要.................................... i
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
誌謝. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
List of Figures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Chapter1 INTRODUCTION........................ 1
1 Overview.......................... 1
2 Motivation.......................... 2
3 Design Principle....................... 3
4 Thesis Organization...................... 4
Chapter2 RELATED WORK........................ 5
1 Traditional Methods...................... 5
1.1 Nearest neighbor interpolation........................ 5
1.2 Bilinear interpolation.................... 6
1.3 Bicubic interpolation.................... 7
2 Deep Learning Methods.................... 8
2.1 SRCNN......................... 8
2.2 VDSR.......................... 9
2.3 LapSRN.......................... 10
2.4 SRResNet.......................... 11
2.4 EDSR.......................... 11
Chapter3 TRAINING DATA USED IN DEEP LEARNING........................ 13
1 Raw Data Collection ........................ 13
2 Training Data Preparation........................ 15
Chapter 4 DEEP LEARNING FOR ACQUIRING SUPER RESOLUTION IMAGE........................ 16
1 Machine Learning and Deep Learning........................ 16
1.1 Machine Learning........................ 16
1.2 Deep Learning........................ 17
2 Layers........................ 18
2.1 Convolution layer........................ 19
2.2 Batch Normalization........................ 20
2.3 Sub-pixel........................ 22
3 Activation Function........................ 22
3.1 Sigmoid........................ 23
3.2 ReLU........................ 24
3.3 LeakyReLU........................ 25
4 Loss Function........................ 25
4.1 Mean absolute error / L1 loss........................ 27
4.2 Mean square error / L2 loss ........................ 28
5 Useful Techniques in Deep Learning........................ 29
5.1 Residual learning........................ 29
5.2 Densely connected convolutional networks........................ 30
5.3 U-Net........................ 31
6 Our Proposed Technique........................ 32
6.1 Our architecture........................ 32
6.2 Training detail........................ 34
6.3 Predictive technology........................ 36
Chapter 5 EXPERIMENTAL RESULTS AND DISCUSSION ........................ 38
1 Expertisesimental Setup........................ 38
2 Evaluation on Different Datasets........................ 39
Chapter 6 CONCLUSION AND FUTURE WORK ........................ 45
1 Conclusion........................ 45
2 Future Work........................ 46
References........................ 47

                                

[1] M.M.Hadhoud,F.AbdEl-Samie,andS.E.El-Khamy,“Newtrendsinhighresolutionimageprocess- ing,” in The Fourth Workshop on Photonics and Its Application, 2004., Giza, Egypt, Egypt, pp. 2–23, May 2004.
[2] C.Dong,C.C.Loy,K.He,etal.,“Imagesuper-resolutionusingdeepconvolutionalnetworks,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 38, pp. 295–307, Feb 2016.
[3] J.Kim,J.K.Lee,andK.M.Lee,“Accurateimagesuper-resolutionusingverydeepconvolutionalnet- works,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, pp. 1646–1654, June 2016.
[4] W. Lai, J. Huang, N. Ahuja, et al., “Fast and accurate image super-resolution with deep laplacian pyramid networks,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1–1, 2018.
[5] C. Ledig, L. Theis, F. Huszár, et al., “Photo-realistic single image super-resolution using a genera- tive adversarial network,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Hawaii, pp. 105–114, July 2017.
[6] B.Lim,S.Son,H.Kim,etal.,“Enhanceddeepresidualnetworksforsingleimagesuper-resolution,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops, Honolulu, Hawaii, pp. 1132–1140, July 2017.
[7] K. He, X. Zhang, S. Ren, et al., “Deep residual learning for image recognition,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, pp. 770–778, 2016.
[8] W. Shi, J. Caballero, F. Huszár, et al., “Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, pp. 1874–1883, June 2016.
[9] G.Huang,Z.Liu,L.v.d.Maaten,etal.,“Denselyconnectedconvolutionalnetworks,”in2017IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Hawaii, pp. 2261–2269, July 2017.
[10] O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image seg- mentation,” in 18th International Conference on Medical Image Computing and Computer-Assisted Intervention, Munich, Germany, 2015.
[11] X.Wang,K.Yu,S.Wu,etal.,“Esrgan:Enhancedsuper-resolutiongenerativeadversarialnetworks,” in The European Conference on Computer Vision Workshops, Munich, Germany, September 2018.
[12] O. Rukundo and H. Cao, “Nearest neighbor value interpolation,” Computing Research Repository, vol. abs/1211.1768, 2012.
[13] R.Timofte,E.Agustsson,L.VanGool,etal.,“Ntire2017challengeonsingleimagesuper-resolution: Methods and results,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition Workshops, pp. 114–125, 2017.
[14] M. Bevilacqua, A. Roumy, C. Guillemot, et al., “Low-Complexity Single-Image Super-Resolution based on Nonnegative Neighbor Embedding,” in Proceedings of the British Machine Vision Confer- ence, Guildford, Surrey, United Kingdom, Sep 2012.
[15] J.-D.Boissonnat,P.Chenin,A.Cohen,etal.,“Curvesandsurfaces,”inProceedingsofthe7thInter- national Conference on Curves and Surfaces, Avignon, France, June 24-30, 2010, Revised Selected Papers, Avignon, France, Springer Science & Business Media, June 2011.
[16] D. Martin, C. Fowlkes, D. Tal, et al., “A database of human segmented natural images and its ap- plication to evaluating segmentation algorithms and measuring ecological statistics,” in Proceedings of the 8th International Conference on Computer Vision, Vancouver, BC, Canada, Canada, vol. 2, pp. 416–423, July 2001.
[17] J.-B. Huang, A. Singh, and N. Ahuja, “Single image super-resolution from transformed self- exemplars,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Boston, Massachusetts, pp. 5197–5206, 2015.
[18] Y.Matsui,K.Ito,Y.Aramaki,etal.,“Sketch-basedmangaretrievalusingmanga109dataset,”Multi- media Tools and Applications, vol. 76, pp. 21811–21838, Oct 2017.
[19] Y.Blau,R.Mechrez,R.Timofte,etal.,“2018PIRMchallengeonperceptualimagesuper-resolution,” Computing Research Repository, vol. abs/1809.07517, 2018.
[20] S.IoffeandC.Szegedy,“Batchnormalization:Acceleratingdeepnetworktrainingbyreducinginter- nal covariate shift,” in ICML’15 Proceedings of the 32nd International Conference on International Conference on Machine Learning, Lille, France, vol. 37, pp. 1097–1105, July 2015.
[21] D. Kingma and J. Ba, “Adam: A method for stochastic optimization,” International Conference on Learning Representations, Banff, Canada, 12 2014.
[22] R. Timofte, V. De Smet, and L. Van Gool, “A+: Adjusted anchored neighborhood regression for fast super-resolution,” in Asian conference on computer vision, Singapore, Singapore, pp. 111–126, Springer, 2014.

全文公開日期 2024/07/31 (校內網路)
全文公開日期 2024/07/31 (校外網路)
全文公開日期 2024/07/31 (國家圖書館：臺灣博碩士論文系統)

簡易檢索 / 詳目顯示

相關論文