視覺特效無疑是電影中不可或缺的技術，自3D電腦繪圖發展以來，產學界致力於創造擬真的效果。然製作擬真特效需要複雜運算與龐大的儲存空間，使電影業者與特效公司背負成本與時間壓力；因此在成本及時程限制下兼顧品質成為電影工業一致的目標。過去特效擬真與否大都由劇組依經驗評估，然視覺效果最終仍取決於觀眾感受，因此每每會發生影片評價與預期產生落差的情形。為此本論文透過圖像實驗探討人眼如何判別視覺特效之逼真度，並分析電腦繪圖與自然影像間的差異及特徵。進而提出超解析訓練影像篩選方法與基於卷積神經網路之視覺特效偵測技術，創新發展一特效品質量測方法來判別視覺特效製作的擬真程度。實驗結果顯示可辨別大部分電腦繪圖與自然影像，並找出不夠擬真的部分，可供特效後製人員檢視其特效創作及算圖品質之客觀檢驗機制。

The quality of visual effects is one of the major factors for the success of a film in box office. Currently the quality of visual effects is evaluated by the practitioners of film post-production and there is very few objective assessments based on the visual experience of audience. This thesis tries to analyze the cues of visual effects in films, and a new quality assessment based on analysis is proposed to evaluate the visual effects. First, the experiments about the fidelity of models, textures and color for computer-generated imagery (CGI) are presented to discuss the quality of visual effects and analyze the differences between CGI and natural images. Then, an assessment method based on deep learning is proposed for visual effects analysis. Experimental results show that the proposed method could identify the CGI and natural images effectively in order to find the non-photorealistic regions. Moreover, the proposed assessment could provide great help to the post-production crew.

論文審定書+i
論文審定書(英文版)+ii
致謝+iii
中文摘要+iv
Abstract+v
Contents+vi
List of Figures+viii
Chapter 1 Introduction+1
1.1 Overview to Computer-Generated Imagery+1
1.2 Motivation+3
1.3 Contributions+4
1.4 Organization+6
Chapter 2 Background Review+7
2.1 CGI in Films+8
2.1.1 Modeling and Texturing+8
2.1.2 Lighting and Rendering+10
2.2 Realistic Visual Effects in Films+12
2.3 Deep Learning+16
2.3.1 Convolutional Neural Networks+18
2.3.1.1 Convolutional Layer+20
2.3.1.2 Pooling Layer+21
Chapter 3 Visual Effects Analysis+23
3.1 The Quality of Visual Effects in Human Vision+24
3.2 The Features of Realistic Visual Effects+26
3.3 The Features of CGI and Natural Images+32
Chapter 4 Visual Effects Detection+34
4.1 Training Image Selection+35
4.2 Feature Extraction and Model Training+38
4.3 Experimental Results+42
4.3.1 Fire Effects I+42
4.3.2 Fire Effects II+44
4.3.3 Smoke Effects I+46
4.3.4 Smoke Effects II+47
Chapter 5 Conclusions and Future Works+49
5.1 Conclusions+49
5.2 Future Works+50
Reference+51

[1] A. K. Moorthy and A. C. Bovik, “Blind image quality assessment: from natural scene statistics to perceptual quality”, IEEE Trans Image Process, vol.20, no.12, pp. 3350-3364, April 2011.
[2] A. Mittal, R. Soundararajan and A. C. Bovik “Making a ‘Completely Blind’ Image Quality Analyzer”, IEEE Signal processing Letters, pp. 209-212, vol. 22, no. 3, March 2013.
[3] M. A. Saad, Alan C. Bovik and Christophe Charrier, “A DCT Statistics-Based Blind Image Quality Index”, IEEE Signal Processing Letters, Vol. 17, no. 6, pp. 583-586, March 2010.
[4] S. Lyu and H. Farid, “How Realistic is Photorealistic?”, IEEE Transactions on Signal Processing, vol. 53, no. 2, pp. 845-850, Feb. 2005.
[5] T. T. Ng, S. F. Chang, J. Hsu, L. Xie and M.P. Tsui, “Physics-motivated features for distinguishing photographic images and computer graphics”, in Proceedings of the 13th annual ACM international conference on Multimedia, pp. 239-248, 2005.
[6] T. Carvalho, H. Farid and E. Kee, “Exposing photo manipulation from user-guided 3D lighting analysis”, Media Watermarking, Security, and Forensics 2015, vol. 9409, March 2015.
[7] J. F. O’Brien and H. Farid, “Exposing photo manipulation with inconsistent reflections”, ACM Transactions on Graphics (TOG), vol. 31, no. 1, Jan. 2012.
[8] Disney explains its powerful new hyperion rendering engine
http://www.cartoonbrew.com/tech/disney-explains-its-powerful-new-hyperion-rendering-engine-117152.html
[9] U. Ghia, K. N. Ghia and C. T. Shin “High-Re solutions for incompressible flow using the Navier-Stokes equations and a multigrid method”, Journal of Computational Physics, vol. 48, no. 3, pp. 387-411, Dec. 1982.
[10] R. Ando, N. Thuerey and C. Wojtan, “A Stream Function Solver for Liquid Simulations”, ACM Transactions on Graphics, vol. 34, no. 4, pp. 8-16, Aug. 2015.
[11] N. Thuerey, C. Wojtan, M. Gross and G. Turk, “multiscale approach to mesh-based surface tension flows”, ACM Transactions on Graphics, vol. 29, no. 4, pp. 1-10, July 2010.
[12] G. E. Hinton and R. R. Salakhutdinov, “Reducing the dimensionality of data with neural networks”, Science, vol. 313, no. 5786, pp. 504-507, July 2006.
[13] J. Schmidhuber, “Deep learning in neural networks: an overview”, Neural Networks, vol. 61, pp. 85–117, Jan. 2015.
[14] Y. Bengio, A. Courville and P. Vincent, “Representation Learning: A Review and New Perspectives”, IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 35, no. 8, pp. 1798-1828, Aug. 2013.
[15] H. Lee, R. Grosse, R. Ranganath and A. Y. Ng, “Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations”, ACM International Conference on Machine Learning, pp. 609-616, June 2009.
[16] Deep learning
https://en.wikipedia.org/wiki/Deep_learning
[17] L. Ma, Z. Lu, L. Shang and H. Li, “Multimodal Convolutional Neural Networks for Matching Image and Sentence”, IEEE International Conference on Computer Vision (ICCV), pp. 2623-2631, Dec. 2015.
[18] F. Meng, Z. Lu, M. Wang, H. Li, W. Jiang and Q. Liu, “Encoding Source Language with Convolutional Neural Network for Machine”, in Proceedings of Conference of the Association for Computational Linguistics(ACL), July 2015.
[19] Backpropagation in convolutional neural network
http://www.slideshare.net/kuwajima/cnnbp
[20] typical CNN architecture
https://commons.wikimedia.org/w/index.php?curid=45679374
[21] Introduction to convolution neural networks
http://infilect.com/2015/08/01/cnn-intro/
[22] K. Nagano, G. Fyffe, O. Alexander, J. Barbič, H. Li, A. Ghosh and P. Debevec, “Skin microstructure deformation with displacement map convolution”, ACM Transactions on Graphics, vol. 34, no. 4, August 2015.
[23] Big Buck Bunny
https://peach.blender.org/download/
[24] Sintel
https://durian.blender.org/