Loading...
|
Please use this identifier to cite or link to this item:
https://nccur.lib.nccu.edu.tw/handle/140.119/153369
|
| Title: | Deepfake與GAN真偽人臉圖像統計分析
Statistical Analysis of Synthetic Images: Deepfake and GAN |
| Authors: | 黃政嘉
Huang, Jia-Jheng |
| Contributors: | 余清祥
陳怡如
Yue, Ching-Syang
Chen, Yi-Ju
黃政嘉
Huang, Jia-Jheng |
| Keywords: | 影像辨識
統計分析
維度縮減
Sobel梯度
資料依賴性
Image Recognition
Statistical Analysis
Dimensionality Reduction
Sobel Gradient
Data Dependency |
| Date: | 2024 |
| Issue Date: | 2024-09-04 14:57:22 (UTC+8) |
| Abstract: | 隨著深度學習技術的發展，人工智慧已可生成的高度逼真的圖像，像是影像深偽影片（Deepfake）和生成對抗網絡（Generative Adversarial Network，GAN）都是知名範例，甚至還有以文字生成影片的模型。這些幾可亂真的偽造影像對資訊安全和個人隱私造成威脅，如何分辨真實、電腦生成影像成為熱門研究議題。有別於深度學習以準確性評估模型，本文希望透過影像資料的特性，配合統計理論及資料分析的概念，作為分辨真實及電腦生成人臉影像的依據。
我們認為Deepfake深偽影像和GANs生成圖像存在局部紋理缺陷，前者呈現過度平滑趨勢，後者則有油畫般的紋理扭曲，這些特性可藉由圖像資料的Sobel梯度計算、一階差分等方法偵測出差異。本文將圖像的RGB紅綠藍三原色等九種色彩空間資料代入上述方法，並以這些資料的統計量（平均數、變異數、離群值）為解釋變數，再使用統計學習、機器學習分類模型，判斷影像是否為電腦生成。分析發現本文提議的統計方法之準確性不亞於深度學習模型，而且使用明顯較少的解釋變數，但需選擇適當的資料切割。以PGGAN影像為例，一階差分的切割數較大時，模型準確率約為95%，Sobel則在切割數為64×64時，模型準確率可達99%。另外，模型準確率有明顯的資料依賴性，尤其是GAN資料集，例如：僅在PGGAN有99%準確率，StyleGAN/StyleGAN2等準確率降至20%左右，但若是比對其他真實資料與生成資料時，準確率可達到90%以上。
With the advancement of deep learning technology, artificial intelligence can generate highly realistic images, such as Deepfake videos and Generative Adversari-al Network (GAN). These technologies pose threats to information security and per-sonal privacy, making the differentiation between real and computer-generated im-ages a critical research topic. This paper aims to distinguish between real and com-puter-generated facial images using image characteristics, statistical theories, and data analysis concepts, rather than deep learning accuracy metrics. We consider that Deepfake and GAN-generated images exhibit distinct local texture defects. Deepfake images show excessive smoothness, while GAN images have painting-like distor-tions. These defects can be detected using methods such as Sobel gradient (Liu et al., 2023) and first-order difference (Chen Huishuang, 2023). This study applies these methods to nine color spaces, including RGB, and uses statistical measures (mean, variance, outlier ratio) as explanatory variables in statistical and machine learning classification models.
Our analysis reveals that the proposed statistical classification method achieves accuracy comparable to deep learning models with fewer explanatory variables, and image split is crucial. For PGGAN images, the first-order difference results show that larger split numbers can achieve 95% accuracy, while the Sobel method reaches 99% accuracy with a split number of 64. Additionally, data dependency significantly impacts model accuracy, particularly in GAN datasets. Using the original training dataset yields better results only for PGGAN, whereas StyleGAN/StyleGAN2 data perform worse. However, cross-validated other GAN datasets achieve over 90% ac-curacy. |
| Reference: | 參考文獻
一、中文文獻
[1]陳慧霜（2023）。「影像分析與深偽影片的偵測」，國立政治大學統計學系學位論文。
[2]李俊毅（2022）。「基於少量訓練樣本下之深偽視訊鑑識技術」，國立屏東科技大學資訊管理系學位論文。
[3]許志仲（2022）。「解決真實世界底下的深偽影視訊偵測問題之研究」，國立成功大學統計學系學位論文。
[4]莊易修（2017）。「成對學習應用於偽造影像/視訊偵測」，國立屏東科技大學資訊管理系學位論文。
[5]李昕（2021）。「融合生成對抗網路及領域知識的分層式影像擴增」，國立中央大學資訊工程學系學位論文。
二、英文文獻
[1]Breiman, L. (2001). “Random Forests”, Machine Learning, 45(1), 5-32. https://doi.org/10.1023/A:1010933404324.
[2]Chen, B., Liu, X., Zheng, Y., Zhao, G., & Shi, Y.-Q. (2022). “A Robust GAN-Generated Face Detection Method Based on Dual-Color Spaces and an Improved Xception”, IEEE Transactions on Circuits and Systems for Video Technology, 32(6).
[3]Cozzolino, D., Gragnaniello, D., Poggi, G., & Verdoliva, L. (2021). “Towards Universal GAN Image Detection”, arXiv preprint arXiv:2112.12606.
[4]Dong, C., Kumar, A., & Liu, E. (2022). “Think Twice before Detecting GAN-generated Fake Images from Their Spectral Domain Imprints”, in Pro-ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
[5]Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Un-ter-thiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., & Houlsby, N. (2021). “An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale”, in International Conference on Learning Repre-sentations (ICLR)
[6]Durall, R., Keuper, M., Pfreundt, F.-J., & Keuper, J. (2020). “Unmasking Deep-Fakes with Simple Features”, arXiv preprint arXiv:1911.00686v3.
[7]Fu, T., Xia, M., & Yang, G. (2022). “Detecting GAN-generated Face Images Via Hybrid Texture and Sensor Noise Based Features”, Multimedia Tools and Ap-plications, 81, 26345–26359.
[8]Fu, Y., Sun, T., Jiang, X., Xu, K., & He, P. (2019). “Robust GAN-Face Detection Based on Dual-Channel CNN Network”, in 2019 12th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)
[9]Giudice, O., Guarnera, L., & Battiato, S. (2021). “Fighting Deepfakes by De-tecting GAN DCT Anomalies”, Journal Imaging 2021, 7(8), 128
[10]Goodfellow, I. J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014). “Generative Adversarial Nets,” Advances in Neural Information Processing Systems, 27.
[11]Gragnaniello, D., Cozzolino, D., Marra, F., Poggi, G., & Verdoliva, L. (2021). “Are GAN Generated Images Easy to Detect? a Critical Analysis of the State-of-the-art”, in Proceedings of the IEEE International Conference on Mul-timedia and Expo (ICME), 1-6.
[12]He, K., Zhang, X., Ren, S., & Sun, J. (2016). “Deep Residual Learning for Im-age Recognition”, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 770-778.
[13]Hu, S., Li, Y., & Lyu, S. (2021). “Exposing GAN-generated Faces Using Incon-sistent Corneal Specular Highlights”, in IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP)
[14]Huang, Y., Juefei-Xu, F., Guo, Q., Liu, Y., & Pu, G. (2022). “FakeLocator: Ro-bust Localization of GAN-Based Face Manipulations”, IEEE Transactions on Information Forensics and Security, 17
[15]Kanopoulos, N., VasanthaVada, N., & Baker, R. L. (1988). “Design of an Image Edge Detection Filter Using the Sobel Operator”, IEEE JOURNAL OF SOL-ID-STATE CIRCUITS, VOL. 23, NO. 2
[16]Karras, T., Aila, T., Laine, S., & Lehtinen, J. (2018). “Progressive Growing of GANs for Improved Quality, Stability, and Variation”, in International Confer-ence on Learning Representations (ICLR)
[17]Karras, T., Laine, S., & Aila, T. (2019). “A Style-based Generator Architecture for Generative Adversarial Networks”, in Proceedings of the IEEE/CVF Con-ference on Computer Vision and Pattern Recognition (CVPR), 4401-4410
[18]Karras, T., Laine, S., Aittala, M., Hellsten, J., Lehtinen, J., & Aila, T. (2020). “Analyzing and Improving the Image Quality of StyleGAN”, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 8110-8119
[19]Li, H., Li, B., Tan, S., & Huang, J. (2020). “Identification of Deep Network Generated Images Using Disparities in Color Components”, Signal Processing, 174.
[20]Li, J., He, H., Huang, Z., & Zhang, C. (2013). “Exposing Computer Generated Images by Using Deep Convolutional Neural Networks”, IEEE Transactions on Information Forensics and Security, 14(5), 1180-1190.
[21]Li, Z., Ye, J., & Shi, Y. Q. (2013). “Distinguishing Computer Graphics from Photographic Images Using Local Binary Patterns”, in Proceedings of the 11th international conference on Digital Forensics and Watermaking
[22]Liu, Y., Wan, Z., Yin, X., Yue, G., Tan, A., & Zheng, Z. (2023). “Detection of GAN generated image using color gradient representation”, J. Vis. Commun. Image R
[23]Liu, Z., Qi, X., & Torr, P. H. S. (2020). “Global Texture Enhancement for Fake Face Detection in the Wild”, in Proceedings of the IEEE Conference on Com-puter Vision and Pattern Recognition (CVPR), 8060-8069.
[24]Liu, Y., & Cheng, L. (2010). “Robust and Fast Blind Image Watermarking Based on DCT dDomain”, Pattern Recognition, 43(5), 1763-1772.
[25]Nguyen, T. T., & Huynh, T. (2012). “A Novel Image Watermarking Scheme Based on Visual Cryptography”, Journal of Visual Communication and Image Representation, 23(7), 1120-1132.
[26]Odena, A., Dumoulin, V., & Olah, C. (2016). “Deconvolution and Checkerboard Artifacts”, Distill. https://doi.org/10.23915/distill.00003.
[27]Qiao, T., Chen, Y., Zhou, X., Shi, R., Shao, H., Shen, K., & Luo, X. (2023). “CSC-Net: Cross-color Spatial Co-occurrence Matrix Network for Detecting Synthesized Fake Images”, IEEE Transactions on Cognitive and Developmental Systems
[28]Pu, J., Mangaokar, N., Wang, B., Reddy, C. K., & Viswanath, B. (2020). “NoiseScope: Detecting Deepfake Images in a Blind Setting”, in Annual Com-puter Security Applications Conference (ACSAC 2020)
[29]Rhee, K. H. (2020). “Detection of Spliced Image Forensics Using Texture Analysis of Median Filter Residual”, IEEE Access, 8, 103374-103384.
[30]Singhal, P., Raj, S., Mathew, J., & Mondal, A. (2022). “Frequency Spectrum with Multi-head Attention for Face Forgery Detection”, in International Con-ference on Neural Information Processing (ICONIP 2022), 200-211
[31]Tan, Y., Liu, S., Huang, Z., & Chen, C. (2023). “Learning on Gradients: Gener-alized artifacts Representation for GAN-generated Images Detection”, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni-tion (CVPR), 1-10.
[32]Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017). “Attention is All You Need”, Advances in Neural Information Processing Systems, 5998-6008.
[33]Wang, S., Wang, X., & Zhang, Y. (2018). “Identification of Natural Images and Computer-generated Graphics Based on Statistical and Textural Features”, Journal of Visual Communication and Image Representation, 55, 495-502.
[34]Yang, X., Li, Y., Qi, H., & Lyu, S. (2019). “Exposing GAN-synthesized Faces using Landmark Locations”, in Proceedings of the ACM International Confer-ence on Multimedia
[35]Zhang, X., Karaman, S., & Chang, S.-F. (2019). “Detecting and Simulating Ar-tifacts in GAN Fake Images (Extended Version)”, in 2019 IEEE Interna-tional Workshop on Information Forensics and Security (WIFS), 1-6 |
| Description: | 碩士
國立政治大學
統計學系
111354027 |
| Source URI: | http://thesis.lib.nccu.edu.tw/record/#G0111354027 |
| Data Type: | thesis |
| Appears in Collections: | [統計學系] 學位論文
|
Files in This Item:
| File |
Description |
Size | Format | |
|---|
| 402701.pdf | | 5519Kb | Adobe PDF | 0 | View/Open |
|
All items in 政大典藏 are protected by copyright, with all rights reserved.
|
