What is Deepfake? How to detect deepfake videos?

Deepfake images and videos are becoming an extremely dangerous trend and are appearing more and more on the internet and in the media. Given the strong rise of this phenomenon, is there any way to detect and prevent its harmful effects?

Translated article by Gavin Phillips, Makeuseof page about how to detect the risks of deepfake technology.

Videos and images appear every day in people's lives, they can exist anywhere from TV, computers, smartphones to online platforms like Youtube, Facebook. However, a technology called deepfake with the ability to "distort the truth" has appeared, making the world increasingly lose faith in the videos and images they see. If one day you or your loved ones unfortunately become a victim of deepfake, what should you do?

Nowadays, the Internet has brought humanity closer together, you can easily know information, images, videos of almost anyone through a few simple operations on social networking sites. Besides the great capabilities that the Internet brings, it also has many security risks. Hackers stealing users' personal data is no longer a strange thing, the most recent event was a few months ago when more than half a million Zoom users' login information was spread by hackers on the dark web in many forms.

In particular, the rise of deepfake in recent years has made the problem of images and videos on the internet more complicated than ever. More seriously, this technology is now no longer a difficult application to use and is also increasingly cheaper.

The rise of deepfakes makes the issue of online images and videos more complex than ever.

What is Deepfake?

Deepfake is a technology that allows users to edit the face and mouth shape of a character in a video or a photo using AI and some open source algorithms for processing.

If deepfake is only used for normal purposes such as teasing friends or relatives, it is a completely harmless technology. But we are talking about the worst cases, deepfake has been misused when it has become a tool largely serving evil actions such as creating celebrity porn videos, revenge porn, fake news.

The term is a portmanteau of 'deep learning' and 'fake .' It's called that because these are two key elements in how deepfakes work. While 'fake' is easy to understand when used to imply that deepfakes' goal is to create fake, different photos than the original, ' deep learning' is more complicated. It involves the 'learning' process of artificial intelligence.

The term Deepfake is a portmanteau of "deep learning" and "fake".

Specifically, every product using artificial intelligence goes through two main steps: loading input data, then building a model and choosing an algorithm to continuously process and learn from those models. Deepfake is no exception, its input data is videos, images, open source algorithms TensorFlow or Keras are all available online. After a period of "learning", AI on deepfake can easily perform the task of merging faces with extremely high similarity.

The dangers of deepfake

Every day, we can see countless images and videos when accessing the internet, especially on social networking sites. I think most of these photos or videos have been edited because in this day and age, platforms like Facebook, Instagram. are no different from each person's face, so almost everyone wants what they post to be beautiful. Thanks to the widespread appearance of image tools from basic like smartphone applications to advanced like Photoshop, users can easily adjust images or videos as they want, even many "photoshop" products are so realistic that they are unrecognizable.

However, deepfake is on a whole other level, it is not an exaggeration to call it a master in creating fake photos and videos.

One of the deepfake videos that shocked viewers around the world was the video that faked the voice and image of former President Obama. Accordingly, the person sitting in front of the camera looked no different from Mr. Obama but spoke in someone else's way. In particular, at the end of the video, the former president said the same words as the famous American actor and director, Jordan Peele, monologuing in one of his videos.

The video fakes the image and voice of Mr. Obama.

While the video is harmless, it raises concerns about the implications of deepfake technology and AI. The barriers to deepfake adoption were initially high, but it's now less of a science now that the data and algorithms are readily available online. Plus, the learning and processing doesn't require supercomputers; instead, a PC with a regular graphics card can easily do the process in a few hours.

Celebrity Face Swap Porn

Taking advantage of the capabilities that deepfake brings, bad guys have used it for personal gain by creating many pornographic products by splicing the faces of famous people such as Emma Watson, Taylor Swift, Scarlett Johansson. Each "adult" film takes place with a familiar script but only the face of this actor is different, attracting tens of millions of views on pornographic sites after being posted. However, no website has taken action to remove or re-upload the original video before.

Marvel beauty is one of the victims of deepfake.

'Celebrity porn videos will continue to exist unless there is a legitimate reason or legal constraint on the sites that post them,' said Giorgio Patrini, CEO and chief research officer at Sensity, a deepfake detection and analytics company  . 'And they will continue to be uploaded without any restrictions.'

Harming women with pornographic films and images

Not only celebrities but also ordinary women have become victims of pornography. A study by Sensity has shown that deepfakes contain a bot operating on the social messaging app Telegram and it is the one who created more than 100,000 nude photos on deepfakes. Of which, the majority of the stolen images are mainly sourced from social networks.

Bots are a perfect complement to deepfake technology, making it easy for users with no knowledge of deepfakes to use. Instead, users only need to do one thing: upload a photo and the bot will automatically take care of the rest. However, according to research, deepfake bots seem to only turn on "active" mode with images of women or users who subscribe to the premium deepfake package (process more photos, sharpen blurry photos).

Bots are the perfect complement to deepfake technology.

The deepfake bot tool quickly caused a wave of fierce protests, people condemned the act of stealing images from Telegram and considered it exploitative, abusive and inhumane. What will happen to the lives of women when those frames are spread online? How will they face their relatives, friends, and colleagues? Not only that, bad guys can take advantage of the spread of images to threaten and blackmail the victim's family or, more seriously, they will become victims in pornographic videos.

There is still an unanswered question as to how a bot that essentially performs such simple and repetitive tasks could have bypassed Telegram's robust system. This is considered almost impossible given that the messaging service is very secure, with high security and the ability to prevent porn bots from entering the system.

"Teasing" the world

Since the video of Obama's image and voice created by deepfake technology was posted, the world has become more and more worried about a similar scenario happening but in a more negative way. The United States is one of the world's leading powers, so the words of its president obviously carry a lot of weight. If deepfake criminals use Obama's image and words to convey a message of war or reaction, it will definitely be a tragedy. Things could even get crazier, the danger of deepfake in this situation is very high.

The fake Obama video event posted in 2018 made many people feel scared about deepfake.

Suppose a deepfake video fakes the content of a major corporation or bank CEO announcing that the stock market has crashed. With the internet, the news will spread globally in a matter of seconds, causing many people to automatically sell their stocks. Any attempt to block or remove the video immediately will be extremely difficult because it is posted publicly online. This is one of the many terrible scenarios that deepfakes bring us.

Not only that, deepfakes are becoming cheaper and easier to use, leading to a large volume of fake videos and images being created. This makes the messages transmitted not only more complex but also more unpredictable. Videos can be about the same person, the same content, but they are built with different tones, locations, styles, etc.

Bending the truth

When we look at deepfake videos and images, we see how sophisticated this technology is. Even when we are warned that it is a fake deepfake video or image, many people are still skeptical and do not believe that it has been edited. To be fair, I think most of us would be like that if it were not for the words "this is fake deepfake data" simply because deepfake images or videos look so real that it is unimaginable.

The "truth distortion" effect of deepfake exists in both positive and negative directions, besides the harm it causes, if we know how to use it properly, we can take advantage of it in many situations. Specifically, if a criminal is arrested and has a video clearly recording his behavior, but this person says: "this is fake evidence, the video is created from deepfake". Instead of continuing to question, let's create deepfake videos to create more evidence that he cannot deny.

The counterfeiting capabilities of deepfakes are extremely sophisticated.

Leadership and social thought acquisition

There have been several documented cases of deepfakes being used to create content that guides and leads people's thoughts by impersonating high-ranking officials. According to information revealed on LinkedIn and Twitter profiles, these videos serve to coordinate activities in strategic government organizations and the high-ranking figures here do not exist but are created using deepfake technology.

However, reputable sources say that the above fake action is actually applied in government networks, national intelligence corporations. with the purpose of collecting information, conducting underground political activities without revealing the identity. However, the risk of deepfake infiltrating and capturing these videos is still unpredictable, we need to prevent it from the beginning.

The risk of deepfakes infiltrating and taking over videos remains unpredictable.

Trust between people disappears

In society, communication is a necessity and occurs frequently and intensively every day. People build relationships with each other through trust and honesty in their words.

This is the weakness that bad guys can exploit through deepfake, they can easily fake someone's words to take advantage of the trust of their relatives, friends, colleagues, from there, carry out countless bad acts such as stealing personal image and video data, sexually harassing, causing family rifts. Such inhumane acts have completely taken place and caused many people to seek death to escape.

How to detect and prevent deepfakes

The fact that deepfake videos and images are becoming more and more sophisticated and perfect has made the ability to recognize them very low. In the early days when this technology was just developing, it still had some weaknesses such as: blurry images, videos with faulty voices and effects. but now deepfake is no longer that simple.

Although there is no specific way to accurately detect deepfakes, in my opinion, if you pay attention to the following 4 signs, you can still recognize deepfake videos and images.

1. Small details:  Although deepfake editing technology has made significant improvements, it is not perfect. Especially for effects related to a series of small details in the video such as hair movement, eye movement, cheek structure, and facial expression when speaking. All of them have not achieved a certain level of perfection and naturalness, especially eye movement. Although deepfake has had tools to adjust this task since its launch, this is still a big flaw in this technology.

Microsoft Video Authenticator is a deepfake detection tool developed by Microsoft.

1. Emotion: Another weakness of deepfake videos is the emotions shown on the characters' faces. This technology can only simulate expressions such as anger, joy, happiness. at an average level and still has the "stiffness" of a machine. Experts say that deepfake is still a long way from achieving the same level of realism as a normal person.
2. Video quality:  In the era of digital technology 4.0, people can record and watch videos at high resolutions such as 4K, HDR 10bit. If a political leader wants to make a statement via video or live stream, one thing is for sure, this process will be filmed in a room with the most advanced equipment. Therefore, if those statements encounter transmission errors, poor sound or image quality, there is a high possibility that deepfake has "involved".
3. Source: The final factor to consider is the source of the deepfake video. Although popular social media platforms always have features to verify user identities to avoid information theft, anything can happen, the system can still have problems. Therefore, when these platforms fail to prevent the spread of videos containing harmful content, this is a sign of deepfake.

Deepfake Detection and Prevention Tools

Currently, many large technology corporations have joined the fight against deepfake, a series of detection tools and platforms to permanently block deepfake have been born.

Typical of these is Microsoft Video Authenticator , a deepfake detection tool developed by Microsoft that can check the authenticity of video data in just a few seconds. Social networking platforms have introduced features to completely block all images and videos edited with deepfakes. But Google is also testing an application to combat audio forgery.