Learning Double-Compression Video Fingerprints Left From Social-Media Platforms

Abstract

Social media and messaging apps have become major communication platforms. Multimedia contents promote improved user engagement and have thus become a very important communication tool. However, fake news and manipulated content can easily go viral, so, being able to verify the source of videos and images as well as to distinguish between native and downloaded content becomes essential. Most of the work performed so far on social media provenance has concentrated on images; in this paper, we propose a CNN architecture that analyzes video content to trace videos back to their social network of origin. The experiments demonstrate that stating platform provenance is possible for videos as well as images with very good accuracy.

Identification of Social-Media Platform of Videos through the Use of Shared Features

“…Table 3 shows the results of this comparison. Splitting the dataset at video level instead of frame level, the method from Amerini et al [39] records a drop in accuracy of 15.47% compared to the configuration used in the original paper.…”

“…Two recent works [8,31] introduced simple yet effective container-based methods to identify video manipulation fingerprints and reconstruct the operating system of the source device, proving the robustness of the method on manipulation introduced by social media platforms. Amerini et al [39] propose a two-stream neural network that analyze I-frames and P-frames in parallel. All frames are preprocessed converting them from RGB to YUV, and the Y-channel of each frame is used as input for the network.…”

“…In this setting the model reaches an accuracy comparable to the single-task learner for the video task. To evaluate the performance of our method, we compared it with the state-of-the-art two-stream network introduced by Amerini et al [39]. To compare the performance of the transfer-learning and multitask-learning-based methods with that of Amerini et al [39], we retrained the model of that work in this new setting.…”

“…Comparison of video patch classification accuracy of our transfer-learning and multitasklearning methods with the one of Amerini et al[39] on the VISION dataset.…”

Identification of Social-Media Platform of Videos through the Use of Shared Features

“…Videos are typically compressed as sequence of groups of pictures (GOP), each of which is made by an alternation of three different kinds of frames: I-frames, which are not derived from any other frame and are independently encoded using a process similar to JPEG compression, and P-frames and B-frames, which are predictively encoded using motion estimation and compensation. While the algorithms used by social platforms are not known, all of these operations leave traces that can be detected Amerini et al ( , 2019 Amerini et al (2021), they can be used to distinguish between distinct social networks. According to the survey by Pasquini et al Pasquini et al (2021), we can identify two main possible steps in the digital life of a media object shared online, namely the acquisition and the upload.…”