In this article, we look at how to spot deepfakes and protect yourself from potential extortion.
467 
One of the most common tasks researchers and journalists face is verifying user videos on social and file-sharing sites such as YouTube, Twitter, and Facebook. There is no foolproof way to check any video, and in some cases it’s almost impossible without access to the source files. However, there are methods that can be applied to verify most material, especially when it is necessary to ensure that breaking news does not contain video footage of older incidents. This article introduces some techniques and begins to educate readers on how to overcome the limitations of existing tools.
In this article, you will find a detailed guide on video verification. It covers advanced methods and tools for authenticating digital video content that are indispensable for journalists, researchers and fact-checkers. The article will help you effectively recognize real and fake videos in the digital age. We hope that after reading this article, readers will not only know how to use the described tools, but also how to creatively approach solving problems so as not to get stuck in them.
The first step in checking videos is the same as for images: a reverse search using services like Google or TinEye. Currently, there are no publicly available tools to search for entire videos in the same way that there are for images, but at least you can search for thumbnails and screenshots. People who distribute fake videos are usually not very inventive and often just post videos that are easy to find and that do not contain obvious signs of contradiction to the events being described, such as superimposed TV credits or an audio track where someone speaks in a language unrelated to the events being described . Therefore, such videos are relatively easy to check for reuse.
There are two ways to search for the source video. The first is to manually take a few screenshots, preferably from the beginning of the video or during key moments, and then upload them to a reverse image search service such as Google Images. The second is to use thumbnails created by a video service, most often YouTube. There is no easy way to determine which frame will be automatically selected for a thumbnail, as Google has developed a complex algorithm for determining the best thumbnails for YouTube videos (you can learn more about this in this article > Google Research Blog). Perhaps the best tool for identifying video thumbnails is Amnesty International’s YouTube DataViewer, which generates the thumbnails used for YouTube videos and allows you to reverse-search those images with one click.
For example, a YouTube user named Action Tube recently posted a video purporting to show a convoy of military equipment in Lithuania, but without citing any original sources. Also, no indication is given as to when this entry was made, so it could have been yesterday or five years ago.
https://www.youtube.com/watch?v=zX7gu_gS3zE
If we paste the link to this video into the search bar of the Amnesty International tool, we will get the exact date and time it was uploaded to Action Tube, as well as 4 thumbnails for reverse image search, which will help us find the original source.
None of the results we get give us a direct indication of the source video; however, some of them, found after the third thumbnail, indicate the single-page reels with which it was shown. If you follow these links, you may not find this thumbnail there, because the videos in the “Next” section on the right side of the YouTube page are selected individually for each user. However, if the video with that thumbnail was there when Google cached the results, you can find it on the saved copy of the page.
Again, none of the videos found are the original source we’re looking for, but when Google cached the pages with them, they also had this thumbnail with a link to the video we were looking for. When we look at the saved copy of the first page of the results list, we see the source of the Action Tube published video with the title “NATO Poland’s Enhanced Forward Presence Battle Group marches under roads to Enhanced Forward Presence Battle Group. Rukli, Lithuania).
We now have all the information we need to find the original footage and confirm that the Action Tube video does indeed show the recent deployment of military equipment in Lithuania. When we enter the title of the video found through the thumbnail search, we get six results. If we sort them by date, we will identify the oldest of them, which became the source for Action Tube.
This brings us to a video that was uploaded by MAJ user Anthony Clas on June 18, 2017 – a day before Action Tube posted the video we reviewed on June 19. This is the same video as on Action Tube.
https://www.youtube.com/watch?v=_4kHuTs1Nog
If we do a little digging on the user who uploaded this video, we can see that he wrote articles for the US Army website on NATO operations in Europe, so he probably works in the public relations department, so thus raising the possibility that his video is the original source of the video posted by Action Tube.
Although reverse image search can detect many fakes, it is still not perfect. For example, the video below, which has 45,000 views, allegedly shows a battle between Ukrainian soldiers and Russian-backed separatists near Svitlodarsk in eastern Ukraine. The video is titled “Battle in the area of the Svitlodar arc in Donbas (filming by the Armed Forces)”. We can see a lot of shooting and artillery fire in it, although the soldiers seem to be laughing during the fight.
By entering the address of this video in the search bar on the Amnesty International website, we will find out the exact date and time it was uploaded and get thumbnails for reverse image search.
Looking at the results, we can see that almost all of the videos found were uploaded around the same time as the one we are looking at, which could lead to the impression that it really shows the battle near Svitlodarsk in December 2016.
Even with a resourceful use of Google’s reverse image search using Amnesty International’s tool, you won’t find the source video, except in articles describing the exposure of fakes that have circulated. For example, if we search for the exact name of the source video (“Caucasus 2012 training night”) with the addition of a screenshot from it, then among the results there will be only fake videos about Svitlodarsk. To understand that this video is a fake, one of two things is necessary: either you have previously encountered the original video, or you have keen eyes (or rather ears) that will tell you that the laughing soldier does not fit into the atmosphere of the battle that is supposedly going on.
And so, what to do? There is no simple answer here – you can only approach the search process creatively. One of the better courses of action would be to try to think like the person who posted the potentially fake video. Thus, in the above example, a soldier laughing can be a clue that perhaps the battle is not real, which raises the question, under what circumstances would a Russian-speaking soldier film what is happening on camera and laugh at the same time. If you wanted to find such a video, what would you look for? You’ll probably want to shoot at night so you can see less detail. You should also look for footage of battles that look impressive, but not the kind that Ukrainians or Russians who follow the war in Donbas would easily recognize. In this way, you may find video recordings of Russian, Ukrainian or Belarusian army exercises, or you can find footage of battles in another country and replace the audio track with a Russian-language recording. If you search for the words “study” and “night”, this video will be the first in the list of results. If after that you did not find the original recording, then to verify the video, it would be best to contact the person who posted it.
Digital tools used to verify materials have limitations and their algorithms can be fooled. People often resort to simple tricks such as mirroring videos, changing the color scheme to black and white, or zooming in or out to avoid detection by reverse image searches. The best way to deal with this is to pay close attention to detail so that certain elements of the environment in the video can be identified as relevant to the event in question.
On September 19, 2016, reports began to spread that the perpetrators of three bombings in New York and New Jersey had been arrested in Linden, New Jersey. Several photos and videos have been shared from various sources, including the following photo of the suspect, Ahmad Khan Rahami, lying on the ground surrounded by police officers.
The exact location in Linden where he was arrested remains unclear, but it’s safe to say that the photos are authentic, given that they show roughly the same location from two different angles. The following video from a local resident also appeared. It is clear that it is real, since it was widely circulated in the media during the day, but how could we quickly verify this in the conditions of breaking news?
From these two photos, we can quickly identify the location of Rahani’s arrest. In the lower left corner of the second photo on the billboard, we can see four numbers (“8211”) and fragments of the words: “-ARS” and “-ODY”. We can also see that the 619 exit is nearby, so we can pinpoint the location more precisely. If we look up a phone number with the digits “8211” in Linden, we get a link to “Fernando’s Auto Sales & Body Work,” which explains the “-ARS” and “-ODY” bits: these are parts of words “cars” and “body” (“cars” and “body”). Additionally, we can find the address for this company: 512 E Elizabeth Ave, Linden, NJ.
Checking this address in Google Street View allows you to quickly make sure that we are on the right track.
In both analyzed photos and videos, the weather is the same: gloomy and wet. At 26 seconds into the video, the driver passes a sign that says “Bower St” and another sign about an intersection with Highway 619, giving us a location that we can match to the one identified in the photos.
Even a cursory glance at Google Maps shows that Bower Street intersects with East Elizabeth Avenue, on which the suspect was arrested near the auto repair shop (marked with a yellow star).
If you have time, by comparing the objects captured on the video with the images from Google Street View, you can determine the exact location where it was captured.
Although each of these steps seems like a lot of work to do, the whole thing should take you no more than five minutes if you know what to look for. If you do not have the opportunity to contact the eyewitness who posted the video footage from the scene, then for verification it will be enough for you to pay attention to the details and “get lost” a little on Google Maps and Google Street View. Video verification should be commonplace not only when reporting an event, but also when you encounter them on social media, as this is one of the fastest ways for fake news to spread.
Compared to photography, video editing requires much more skill and effort so that after adding or removing certain elements from the frame, it continues to look natural. In many cases, videos are altered not to fool fact-checkers, but to avoid detection by algorithms that look for copyrighted material. For example, movies, TV shows and sporting events can be viewed without breaking copyright laws (albeit in a somewhat unusual way) using mirroring when uploaded to YouTube. The best way to quickly determine if a video is mirrored is to look for letters and numbers.
The screenshot below shows how footage of a terrorist attack at Moscow’s Domodedovo Airport in 2011 was appropriated as footage of terrorist attacks at Brussels and Istanbul airports. Techniques used by the forgers included enlarging parts of the footage, adding fake timestamps and changing the color tone to black and white. Also, they often place a flashy logo on top of the image to make it even more difficult to reverse engineer.
There is no easy way to identify such fakes with tools. You have to rely on common sense and be creative in your search. As with the recent misappropriation of Russian military exercises as a combat video, it must be taken into account that a faker will look for a source for the video. A search for “airport explosion” or “terrorist attack surveillance camera” will lead to a video of the terrorist attack at Domodedovo airport and will return results much faster than a reverse Google image search using screenshots.
While many believe that technological advances will eradicate fake news and fake content in the future, there is currently no way to identify fake videos and verify content with a high degree of accuracy in the digital world. In other words, in addition to strict control over the content of materials posted on social networks and YouTube, developers are losing the arms race to even the most creative fakes. Digital tools are important for checking fake news, but creativity is more important.
467 
390 
380