BiHPF: Bilateral High-Pass Filters for Robust Deepfake Detection

Abstract

The advancement in numerous generative models has a two-fold effect: a simple and easy generation of realistic synthesized images, but also an increased risk of malicious abuse of those images. Thus, it is important to develop a generalized detector for synthesized images of any GAN model or object category, including those unseen during the training phase. However, the conventional methods heavily depend on the training settings, which cause a dramatic decline in performance when tested with unknown domains. To resolve the issue and obtain a generalized detection ability, we propose Bilateral High-Pass Filters (BiHPF), which amplify the effect of the frequency-level artifacts that are known to be found in the synthesized images of generative models. Numerous experimental results validate that our method outperforms other state-of-the-art methods, even when tested with unseen domains.

FrePGAN: Robust Deepfake Detection Using Frequency-Level Perturbations

“…Also, (Frank et al 2020) analyze the frequency artifacts using Discrete Cosine Transform, while (Zhang, Karaman, and Chang 2019) exploit the artifacts induced by the up-sampler of GANs. Others (Durall, Keuper, and Keuper 2020;Durall et al 2019) exploit the spectral distortions via azimuthal integration, while (Jeong et al 2021) adopt the bilateral high-pass filters for generalized detection. (He et al 2021) propose to re-synthesize testing images and extract visual cues for flexible detection.…”

“…. To validate the effectiveness of the proposed deepfake detector, we select the image-based(Wang et al 2020) and frequencybased state-of-the-art models(Frank et al 2020;Durall, Keuper, and Keuper 2020;Jeong et al 2021) for comparison.Deepfake Detection of Unknown Categories As shown in Figure4, we conduct various experiments using the three classifiers to analyze the performance of the models with 20 unknown categories. We compare our model's performance to the previous state-of-the-art models(Wang et al 2020;Durall, Keuper, and Keuper 2020;Frank et al 2020).…”

FrePGAN: Robust Deepfake Detection Using Frequency-level Perturbations

“…Also, (Frank et al 2020) analyze the frequency artifacts using Discrete Cosine Transform, while (Zhang, Karaman, and Chang 2019) exploit the artifacts induced by the up-sampler of GANs. Others (Durall, Keuper, and Keuper 2020;Durall et al 2019) exploit the spectral distortions via azimuthal integration, while (Jeong et al 2021) adopt the bilateral high-pass filters for generalized detection. Recently, (He et al 2021) propose to resynthesize testing images and extract visual cues for flexible detection.…”

“…Ablation Test with Various Settings. select the image-based(Wang et al 2020) and frequencybased state-of-the-art models(Frank et al 2020;Durall, Keuper, and Keuper 2020;Jeong et al 2021) for comparison.Deepfake Detection of Unknown CategoriesAs shown in Figure4, we conduct various experiments using the three classifiers to analyze the performance of the models with 20 unknown categories. We compare our model's perfor-mance to the previous state-of-the-art models(Wang et al 2020;Durall, Keuper, and Keuper 2020;Frank et al 2020).…”