隨著科技日益進步，影像編輯軟體也越來越發達。使用者可以透過修圖軟體，輕鬆地更改影像資訊，且修改後的內容，僅憑肉眼無法分辨出真假；另外，由於機器學習的發達，電腦亦具備自動生成影像的能力，可以輕易地產生出實際上並不存在的內容，而讓人類無法察覺不合理之處，若此項技術遭受有心人士不當使用，將會造成嚴重的社會問題。
未經授權就修改資訊的行為，我們稱為變造 (forgery) 或是竄改 (tampering)。通常會受到竄改的資訊，不外乎是文字或是圖片；相較於影像，文字的變造比較簡單，只需要改變或移動文字即可，而影像的變造，通常以合成居多。至於是否能被肉眼識破，完全依靠變造者的技術；但是，近年來，由於深度學習技術大爆發，使得傳統的偽造影像偵測方式無法使用。這是因為現今的偽造影像完全由電腦生成，而實際上並不存在，遺憾的是，非編輯製作的偽造圖像不會留下任何篡改痕跡。
使用生成對抗網絡(Generative Adversarial Network; GAN)是目前電腦生成影像技術中最常用的方法。生成對抗網路包含一個生成器與一個判別器；生成器的目標是生成出接近真實樣本的影像，而判別器的目標是將生成影像從真實樣本中區分出來；若判別器可以區分真實影像和偽造影像，則調整生成器的參數，直到判別器無法辨識偽造影像為止。
本論文所提之方法分成兩個部分。首先，我們針對生成對抗網路所生成的真、偽影像分別做離散傅立葉變換；接著，將轉換後取得的頻譜影像，輸入到深度神經網路進行模型訓練。為了提升模型的辨識性能，我們納入了對比式學習(Contrastive Learning)，使電腦直接學習真、偽影像的差別。於實驗的部分，我們選擇了用三種不同的生成對抗網路，稱為 DCGAN、CycleGAN 和 AutoGAN，來產生偽造影像。透過我們提出的方法來辨識此三種不同生成對抗網路的偽造影像，實驗結果表明，使用我們提出的方法來檢測三種不同的 GAN 影像，平均準確率達到99.10%，與訓練和檢測特定目標相比，我們的方法可以更廣泛地識別從不同來源生成的偽造影像。

With the advancement of science and technology, image editing software has become more developed. The user can easily change the image information through the retouching software. And the revised content cannot be distinguished from the true and false with the naked eye. In addition, due to machine learning advancement, computers can automatically generate images, which can quickly produce content that does not exist, so that humans cannot detect unreasonable places. If the man having evil intentions improperly uses this technology, it will cause serious social problems.
The act of modifying information without authorization is called forgery or tampering. Information usually subject to tampering is nothing more than text or pictures. Compared with images, text modification is relatively simple, since people only need to change or move the text. The forged images are usually done by synthesis. Whether a forged image can be seen through with the naked eye entirely depends on the skills of the forgery maker. However, owing to the explosion of deep learning technology in recent years, we cannot use traditional methods of detecting forged images. Because nowadays, forged images are fully generated by computers and do not exist. Unfortunately, the forged images that are not produced by editing will not remain any traces of tampering.
Employing Generative Adversarial Network (GAN) is the mostly common used method in computer-generated image technology currently. The generative adversarial network includes a generator and a discriminator. The goal of the generator is to generate images close to actual samples, whereas the goal of the discriminator is to distinguish the generated image from the actual sample. If the discriminator can distinguish between real and forged images, then adjust the parameters of the generator until the discriminator cannot identify a forged image.
The method proposed in this thesis is divided into two parts. First of all, we separately do discrete Fourier transform for both the real and forged images generated from the generative adversarial networks. Then the spectrum images obtained from feature extraction are fed to deep neural networks for model training. In order to enhance the detection capability of the model, we incorporate contrastive learning to make the model directly learns the difference between real and forged images. We choose three different generative adversarial networks, namely DCGAN, CycleGAN, and AutoGAN, to generate forged images for testing our proposed method. The experimental results reveal that the average accuracy rate reaches 99.10% using our proposed method to detect the three different GAN images. Compared with training and detecting specific targets, our method can more widely detect forged images generated from different sources.

[1] I. Goodfellow et al., “Generative adversarial nets,” in Proceedings of the 27th International Conference on the Neural Information Processing Systems, Cambridge, Massachusetts, vol. 2, 2014, pp. 2672-2680.
[2] A. Radford, L. Metz, and S. Chintala, “Unsupervised representation learning with deep convolutional generative adversarial networks,” arXiv preprint arXiv:1511.06434, 2016.
[3] T. Karras et al., “Progressive growing of GANs for improved quality, stability, and variation,” arXiv preprint arXiv:1710.10196, 2017.
[4] T. Karras, S. Laine, and T. Aila, “A style-based generator architecture for generative adversarial networks,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, California, 2019, pp. 4401-4410.
[5] A. Brock, J. Donahue, and K. Simonyan, “Large scale GAN training for high fidelity natural image synthesis,” in Proceedings of the International Conference on Learning Representations, New Orleans, Louisiana, 2019.
[6] J.-Y. Zhu et al., “Unpaired image-to-image translation using cycle-consistent adversarial networks,” in Proceedings of the IEEE International Conference on Computer Vision, Venice, Italy, 2017, pp. 2223-2232.
[7] Y. Choi et al., “Stargan: Unified generative adversarial networks for multi-domain image-to-image translation,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, Utah, 2018, pp. 8789-8797.
[8] P. Isola et al., “Image-to-image translation with conditional adversarial networks,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Hawaii, 2017, pp. 1125-1134.
[9] F. Marra et al., “Detection of GAN-generated fake images over social networks,” in Proceedings of the IEEE Conference on Multimedia Information Processing and Retrieval, Miami, Florida, 2018, pp. 384-389.
[10] D. Cozzolino, D. Gragnaniello, and L. Verdoliva, “Image forgery detection through residual-based local descriptors and block-matching,” in Proceedings of the IEEE International Conference on Image Processing, Paris, France, 2014, pp. 5297-5301.
[11] D. Cozzolino, G. Poggi, and L. Verdoliva, “Recasting residual-based local descriptors as convolutional neural networks: an application to image forgery detection,” in Proceedings of the 5th ACM Workshop on Information Hiding and Multimedia Security, New York, 2017, pp. 159-164.
[12] B. Bayar and M. C. Stamm, “A deep learning approach to universal image manipulation detection using a new convolutional layer,” in Proceedings of the 4th ACM Workshop on Information Hiding and Multimedia Security, Vigo, Spain, 2016, pp. 5-10.
[13] N. Rahmouni et al., “Distinguishing computer graphics from natural images using convolution neural networks,” in Proceedings of the IEEE Workshop on Information Forensics and Security, Rennes, France, 2017, pp. 1-6.
[14] G. Huang et al., “Densely connected convolutional networks,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Hawaii, 2017, pp. 4700-4708.
[15] C. Szegedy et al., “Rethinking the inception architecture for computer vision,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, 2016, pp. 2818-2826.
[16] F. Chollet, “Xception: Deep learning with depthwise separable convolutions,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Hawaii, 2017, pp. 1251-1258.
[17] J. Fridrich and J. Kodovsky, “Rich models for steganalysis of digital images,” IEEE Transactions on Information Forensics and Security, vol. 7, no. 3, pp. 868-882, 2012.
[18] C. Szegedy et al., “Going deeper with convolutions,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Boston, Massachusetts, 2015, pp. 1-9.
[19] X. Zhang, S. Karaman, and S.-F. Chang, “Detecting and simulating artifacts in GAN fake images,” in Proceedings of the IEEE International Workshop on Information Forensics and Security, Delft, Netherlands, 2019, pp. 1-6.
[20] E. Denton et al., “Deep generative image models using a laplacian pyramid of adversarial networks,” arXiv preprint arXiv:1506.05751, 2015.
[21] J. T. Springenberg et al., “Striving for simplicity: The all convolutional net,” arXiv preprint arXiv:1412.6806, 2014.
[22] T. Salimans et al., “Improved techniques for training GANs,” Advances in Neural Information Processing Systems, vol. 29, pp. 2234-2242, 2016.
[23] M. Heusel et al., “GANs trained by a two time-scale update rule converge to a local nash equilibrium,” Advances in neural information processing systems, vol. 30, pp. 6626-6637, 2017.
[24] S. Barratt and R. Sharma, “A note on the inception score,” arXiv preprint arXiv:1801.01973, 2018.
[25] C.-C. Hsu, Y.-X. Zhuang, and C.-Y. Lee, “Deep fake image detection based on pairwise learning,” Applied Sciences, vol. 10, no. 1, pp. 370-383, 2020.
[26] K. He et al., “Deep residual learning for image recognition,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, 2016, pp. 770-778.
[27] J. Bromley et al., “Signature verification using a “siamese” time delay neural network,” International Journal of Pattern Recognition and Artificial Intelligence, vol. 7, no. 04, pp. 669-688, 1993.
[28] K. Qiu et al., “Siamese-ResNet: Implementing loop closure detection based on siamese network,” in Proceedings of the IEEE Intelligent Vehicles Symposium, Changshu, China, 2018, pp. 716-721.
[29] R. Hadsell, S. Chopra, and Y. LeCun, “Dimensionality reduction by learning an invariant mapping,” in Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, New York, 2006, pp. 1735-1742.
[30] T. Park. “Cyclegan datasets,” [Online]. Available: https://people.eecs.berkeley.ed u/~taesung_park/CycleGAN/datasets/. [Accessed July 30, 2020].
[31] J. Deng et al., “Imagenet: a large-scale hierarchical image database,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Miami, Florida, 2009, pp. 248-255.
[32] R. Tyleček and R. Šára, “Spatial pattern templates for recognition of objects with regular structure,” in Proceedings of the German Conference on Pattern Recognition, Saarbrücken, Germany, 2013, pp. 364-374.
[33] M. Cordts et al., “The cityscapes dataset for semantic urban scene understanding,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Nevada, 2016, pp. 3213-3223.