In recent years, the field of digital imaging has witnessed impressive growth to the extent that every smartphone now incorporates a video camera for recording videos of good quality, at no cost, and without any constraints. On the other hand, the fast growing Internet technology has substantially contributed to the widespread use of digital videos via web-based multimedia systems and smartphone mobile applications such as YouTube, FaceBook, Twitter, WhatsApp, etc. However, as the recording and distribution of digital videos have become affordable nowadays, security challenges have threateningly emerged and spread worldwide. The most obvious challenging issues that normally lead to digital media conflicts include video authenticity, illegal video copying/distribution and malicious video tampering. For instance, the giant Qatari channel Being Sport has been the victim of an unprecedented illegal copying operation led by BeoutQ, a piracy channel that has streamed live coverage of all 64 world cup football games in 2018 as well as the opening games of the Premier League and Ligue 1 over the current season (2018/2019) [1][2].
Also, in 2017 the Qatar News Agency (QNA) website was hacked by an unknown entity. A false statement attributed to the Emir of Qatar Sheikh Tamim Bin Hamad Al Thani while attending a graduation ceremony for Qataris doing national service, was published. The Qatari government said the state TV footage posted online was "fake videos" and authorities had launched an investigation [3]. As such, digital videos can be used to testify incidents and provide legally acceptable evidence for courtroom purposes. It is essential to gather genuine information about the video for forensic investigations.
This research proposal aims to develop a versatile platform for processing and analyzing digital smartphone videos in order to identify the smartphone used to record the input video as well as the brand and model of the smartphone. It also aims to authenticate the video contents and detect forgeries in maliciously manipulated videos. In particular, we are concerned with the following:
• Originality and integrity: The platform verifies whether a smartphone video is genuine and authentic. If the video has been maliciously tampered with, we aim to detect the forgery and localize it.
• Source identification and metadata extraction: The platform identifies the source smartphone. If many smartphone devices of the same model and make happen to be under investigation, we aim to identify the right device, which was used to record the videos under analysis. The platform should also be able to determine whether two given videos are recorded by the same smartphone (i.e., device linking).
This will advance the field of forensic science, with the potential for applications in high profile cases that require the extraction of evidential information for courtroom purposes. We will construct a public smartphone video dataset for source identification and content authentication. We also propose to estimate sensor pattern noise (SPN), which is a fingerprint of the device, for source smartphone identification with robust transform-based image filtering techniques and weighted averaging that can take advantage of the multiple frames in each video to cancel out the undesirable types of noise. Pre-processing methods will be applied to facilitate the accurate estimation of SPN. Machine learning techniques will be adopted for the classification of smartphone brands and makes. Finally, SPN will be the key to video content authentication and forgery detection with a focus on spatial localization of malicious manipulations. The research will be carried out by a specialized team in machine learning and computer vision from Qatar University and Northumbria University.