Deepfakes: A new threat to personal privacy and identity
By Freddie Lichtenstein
Video has never been more easily created, accessible, and abundant.
Mobile phones continue to develop greater video recording quality, better internet speeds enable easier video sharing, and social media platforms continue to prioritise video viewing to improve engagement.
In 2022, Cisco estimates that 82% of all online traffic stems from video streaming and downloads, and this trend is only increasing.
However, a significant threat grows alongside this trend; Now more than ever, online harmful video imagery is being created and shared online.
So how are online intelligence investigators fighting back?
Video forensics enables online investigators to source more actionable intelligence to identify criminals and safeguard victims.
Discover more about the unique applications of video forensics below.
Videos are essentially a large collection of images and so every frame may contain useful intelligence that investigators can use to further their research. Due to video compression capabilities, a very high-quality video can now be a very small file that can easily be uploaded and shared.
For investigators, that means the potential for clear video frames that can be effectively analysed.
As well as containing more data, video forensics provides unique ways to source new insights.
Due to the nature of videos, investigators can detect, identify, and capitalise on the inclusion of audio prompts. These can be both vocal (a subject in or out of frame’s speech) or environmental (stemming from a radio, street noise, or other distinctive sources).
Audio prompts may provide additional data that is critical to investigations. A snippet of radio noise may lead to a timestamp, or vocal cues can help refine searches by dialect, accent, language, or other cues.
Electrical Network Frequency (ENF) analysis also combines specific audio indicators with other datasets.
With acute analysis tools, researchers can use the background hum of mains power in videos to source a timestamp to the second.
In the UK, the frequency of the electricity supplied through the national grid is 50 hertz. You can hear the hum of it in daily life, such as when you partially plug speakers in. Even when you can’t hear it, digital recordings will often pick up this hum. It can be found in lights, pylons, plug sockets and more. Even if you can’t hear it, recordings can pick it up at a very low level.
However, over time the national grid experiences extremely delicate fluctuations.
By comparing the fluctuations in videos to data collected, either manually or through the national grid’s historic frequency datasets, analysts find an accurate timestamp. They can also ensure that videos are genuine and free from editing that may compromise authenticity. This is particularly important in an age of deepfakes and fake news.
Although few are aware about ENF, it’s not a new area of research. The Met police were harnessing ENF video analysis in the courtroom back in 2012.
Video forensics can provide further insights from visual cues. Bellingcat demonstrated just this when they determined the time of day via the shadows found in video imagery. With access to this data, users can use tools such as SunCalc to geolocate video imagery. Learn more in Bellingcat’s case study.
While video forensics can bring a range of additional information to investigators, some complex challenges must be navigated.
Video forensics requires significantly more data processing resources than traditional image forensics, due to the larger volume of data ingested. This can also be strenuous on data storage and analysis, becoming more and more time-consuming as a result.
As videos continue to be redistributed online, they can get endlessly recompressed. This can damage the resolution of frames and potentially remove valuable audio and visual indicators in the process.
Perpetrators may also deliberately edit videos to prevent comparison or identification.
These challenges are difficult hurdles to overcome, yet our work in R&D is attempting just that with regular video searching projects. We’re always looking to improve the functionality of the CameraForensics platform.
While previous we have focused on demonstrating the ability to compare duplicate video recognition software, we’re acutely aware of the possibilities of video forensics moving forward.
If you’re a Law Enforcement Agency or online intelligence investigator, we want to hear from you. What are your thoughts on the possibilities of video forensics, and what capabilities you see as being the highest priority when it comes to enabling video forensics tools? Get in touch with us to let us know.