Search

Deep learning algorithms have shown great performance in multimedia forensics applications using supervised learning on large-scale labeled datasets. However, constructing such extensive labeled datasets can be challenging and costly in several multimedia forensics scenarios. Additionally, heavyweight deep learning models with complex architectures and a large number of parameters require significant hardware resources for training. To address these challenges in the context of image cropping detection, a common multimedia forensics application, we propose a semi-supervised deep learning framework capable of training on a large amount of unlabeled image samples. In this framework, we leverage a teacher model, trained on a small set of labeled image samples, to rank the confidence scores of image samples in a large-scale unlabeled dataset. By utilizing the ranked image samples, we train a student network successfully. To validate the effectiveness of our collaborative training framework across various image cropping detection scenarios, we conduct extensive experiments on a large-scale dataset. The experimental results clearly demonstrate that our semi-supervised learning approach achieved a state-of-the-art performance compared to existing supervised detection frameworks, achieving an accuracy of 91.79% on the BOSSbase dataset and 89.23% on the Alaska dataset. Furthermore, we conducted in-depth research on various factors that influence detection performance in the context of semi-supervised learning. These factors include pairings of teacher–student models, the top-K selection approach, the number of unlabeled samples, the number of iterations in self-training, and the proportion of high-confidence samples using in semi-supervised learning. [ABSTRACT FROM AUTHOR]

Data hiding is an essential practice for ensuring the security of confidential data and secret communication and thus has been intensively studied. The development of deep learning enables the realization of video hiding, which aims to hide a video into another. Different from image data hiding, stego videos generated by video hiding methods need to undergo video coding in real-world scenarios because the unencoded raw videos are too large to be easily transmitted or stored. However, previous video-hiding works did not consider this, leading to a common vulnerability to information loss caused by video coding on stego videos. In other words, previous works considered videos simply as sequences of temporally continuous images, and thus the image sequences output from the data hiding network were fed directly into the recovery network to obtain secret information without video coding. We propose a module called CompressionSimulator to make stego videos robust to video coding. CompressionSimulator is pre-trained to simulate the distortion caused by video coding, and then trained together with the whole framework to make the data hiding network robust to video coding. Furthermore, an invertible neural network based framework is introduced for our video hiding task, i.e. hiding one video in another, in which we design an auxiliary module to capture the spatio-temporal relationships of videos. In general, we develop a video hiding network (VHNet) that provides video coding robustness. We conduct experiments to prove the necessity of incorporating video coding into the design of video hiding networks. Results of experiments conducted on the UCF 101 dataset show that our proposed work outperforms previous methods, and our framework achieves optimal performance in both visual and quantitative comparisons. To the best of our knowledge, we are the first paper to propose a learning-based scheme to improve the robustness of hiding videos with videos. [Display omitted] • Consider real-world applications and design a module to enhance robustness against video coding. • Use an invertible neural network to realize large-capacity data hiding. • First attempt to realize video hiding with robustness to video coding. [ABSTRACT FROM AUTHOR]