A New Dataset for Forged Smartphone Videos Detection: Description and Analysis

The advancement of Internet technology has significantly impacted daily life, which is influenced by digital videos taken with smartphones as the most popular type of multimedia. These digital videos are extensively sent through various social media websites such as WhatsApp, Instagram, Facebook, Twitter, and YouTube. The development of intelligent and simple editing tools has favoured the transformation of multimedia content on the Internet. As a result, these digital videos' credibility, reliability, and integrity have become critical concerns. This paper presents a video forgery (Copy-move forgery) dataset in which 250 original videos are manipulated mainly by two forgery techniques: Insertion and Deletion. Inserting transparent objects into the original video without raising suspicion is one type of manipulation performed. Another type of forgery presented on the dataset is the removal of objects from the original video without notifying the viewers. The videos were collected from five different mobile devices, namely, IPhone 8 Plus, Nokia 5.4, Samsung A50, Xiomi Redmi Note 9 Pro and Huawei Y9-1. The forged videos were created using a popular video editing software called Adobe After Effects as well as usage of other software such as Adobe Photoshop and AfterCodecs. Since the source of the videos is known, PRNU-based methods can be suitable for applying to the dataset. Experiments were performed using classical and deep learning methods. The results are recorded and discussed in detail, showing that improvements are essential for the dataset. Furthermore, the forged videos of this dataset are comparatively large when compared to other datasets that performed copy-move forgery.

Automated summary

The advancement of Internet technology has greatly impacted daily life, with digital videos from smartphones being the most popular form of multimedia. These videos are extensively shared on various social media platforms. The credibility, reliability, and integrity of these videos have become major concerns. This paper presents a dataset of video forgeries, specifically copy-move forgeries, where objects are inserted or deleted from original videos without detection. The dataset includes videos from five different mobile devices and experiments are conducted using classical and deep learning methods.