As deepfake becomes more sophisticated, the demand for fake facial image detection is increasing. Although great progress has been made in deepfake detection, the performance of most existing deepfake detection methods degrade significantly when these methods are applied to detect low-quality images for the disappearance of key clues during the compression process. In this work, we mine frequency domain and RGB domain information to specifically improve the detection of low-quality compressed deepfake images. Our method consists of two modules: (1) a preprocessing module and (2) a classification module. In the preprocessing module, we utilize the Haar wavelet transform and residual calculation to obtain the mid-high frequency joint information and fuse the frequency map with the RGB input. In the classification module, the image obtained by concatenation is fed to the convolutional neural network for classification. Because of the combination of RGB and frequency domain, the robustness of the model has been greatly improved. Our extensive experimental results demonstrate that our approach can not only achieve excellent performance when detecting low-quality compressed deepfake images, but also maintain great performance with high-quality images.
Source camera identification is an important branch in the field of digital forensics. Most existing works are based on the assumption that the number of training samples is sufficient. However, in practice, it is unrealistic to obtain a large amount of labeled samples. Therefore, in order to solve the problem of low accuracy for existing methods in a few-shot scenario, we propose a novel identification method called prototype construction with ensemble projection (PCEP). In this work, we extract a variety of features from few-shot datasets to obtain rich prior information. Then, we introduce semi-supervised learning to complete the construction of prototype sets. Subsequently, we use the prototype sets to retrain SVM classifiers, and take the posterior probability of each image sample belonging to each class as the final projection vector. Finally, we obtain classification results through ensemble learning voting. The PCEP method combines feature extraction, feature projection, classifier training and ensemble learning into a unified framework, which makes full use of image information of few-shot datasets. We conduct comprehensive experiments on multiple benchmark databases (i.e., Dresden, VISION and SOCRatES), and empirically show that our method achieves satisfactory performance and outperforms many recent methods in a few-shot scenario.