An AI face-changing app suddenly hits the whole network! With "only one photo, starring in the world's best show", it quickly harvested millions of downloads, and became the 7th place in the App Store ranking. Users only need to upload a photo in the APP to replace their face with the protagonist image in the film and television drama, which can almost be fake.
This carnival for ordinary users to experience the powerful capabilities of AI at close range has only lasted less than half a day and then took a turn for the worse. In this era where facial recognition is needed everywhere, the panic that “face” is at risk of being leaked extinguished all enthusiasm for embracing technology.
"My first instinct is that the chain reaction caused by AI face-swap technology is hugely risky. Face biometrics, which is widely used in the financial field, and face-swiping payment will be risky. Face-changing videos are also used as evidence of forgery. It may be exploited by illegal elements. These consequences are unimaginable.” As the CTO of Sumei Technology, who has been deeply involved in business risk control for four years, Liang Kun expressed his opinion in the work group when he saw the application of this technology. Worry.
After 24 hours, Liang Kun announced that SUMEI Technology's face-changing recognition product was officially launched. "Business risk control is our professional field. It is our mission to develop recognition products for risky behavior as soon as possible."
"Out of control" AI face changing technology
The core of AI face-changing technology is an "auto-encoder", this "auto-encoder" is actually a deep neural network, it can receive data input, and compress it into a small code, and then re-encode from this code Generate raw input data.
Face-swap technology is not a new technology, it has long been in the production of film and television dramas. However, face-changing in early film and television productions required long and complicated work by editors and CG (computer three-dimensional animation) experts, so it did not have the basis for popularization.
However, it has not been long since AI changed faces. In December 2017, a person named Deep Fakes replaced the heroine in the erotic movie with the Hollywood star Gal Gadot. This is the first time people have seen it publicly. This technology to AI face change.
The source behind the “Deep Fakes” technology is an AI model called GAN (Generative Adversarial Network), which was established in October 2014 by a GAN paper published by the former Google’s famous god-level technology expert Ian Good Fellow Given its status, GAN is currently one of the most popular research directions in the field of artificial intelligence.
However, this technology quickly got out of control. A large number of victims filed complaints.
Even Facebook CEO Mark Zuckerberg became the victim. One of his tampered videos surfaced in which he questioned the company's user data privacy.
The list of victims keeps growing. Many Hollywood actresses, including Emma Watson and Scarlett Johansson, lie down with guns and are changed faces into adult movie heroines, falling into the "Naughty Photos" incident for no reason. Some start-ups, US government agencies, and academia are fighting the so-called "Deep Fakes". They worry that tampered videos and photos will be used to disrupt the US presidential election next year.
"Face" is not safe and money may not be safe
Will the face-swap technology be used to produce fake videos, such as speeches pretending to be public figures, forging evidence in litigation, forging ambiguous and even adultery videos for extortion, forging non-existent evidence, and forging identity for online fraud?
In addition to a series of worries about fake videos, ordinary Chinese people worry the most that if "face" is not safe, "money" will also face great risks. After all, China's Internet payment is far ahead of other countries in the world.
At present, most banks and other financial institutions have opened face recognition login APP functions, and scenarios such as "face-swiping" payment and even remote signing are becoming more common. Some netizens on Weibo are worried: "Illegal collection of facial information", "In case the database is taken away by hackers someday, your Alipay can be stolen in minutes, just watch it yourself."
The two major Internet payment giants in China responded quickly. "Face payment" uses 3D face recognition technology. Before face recognition is performed, detection is performed through a combination of software and hardware to determine the collection. Whether your face is generated by photos, videos or software simulations, it can effectively avoid the fraudulent use of identity caused by various face forgeries. Alipay said that even in the unlikely event of fraudulent use of the account, Alipay will make full payment through the insurance company.
However, the core scene of facial payment is the facial payment in major shopping malls. In this case, you don’t need to touch the APP, and simply face the camera. Just use the video to make the camera think that it is a real person. Is it possible to avoid stolen payment? question mark.
AI face recognition and technology to defend against risks
According to media reports, the current AI face-changing black production has formed a complete industrial chain in China, from downstream finished erotic videos, midstream customized videos to upstream software and tutorials, and some sellers even provide products and after-sales services for the entire industry chain. .
"Risk control technology must respond quickly to prevent user losses caused by risky behaviors." With this sense of mission, CTO Liang Kun quickly mobilized the team and immediately launched AI face-changing recognition products to identify which videos use AI Change face.
"The AI face recognition products of SUMEI Technology are based on face detection technology, face feature extraction technology, and the XceptionNet network optimized for face recognition scenes. Through comprehensive video multi-frame recognition, it usually only takes 10 seconds to include the target. The video of the character can quickly and accurately identify whether the video content is forged by AI face-changing technology." Liang Kun said.
The AI face-changing recognition model of Sumei Technology can automatically extract important features such as face key points, face edges, face orientation, etc. These features help the model to achieve more than 90% accuracy in video recognition, effectively preventing various uses of AI The face-changing video implemented fraudulent behavior.
In Liang Kun’s view, as a solution expert in the field of artificial intelligence, Shumei Technology will continue to be committed to giving full play to its professional advantages to help customers defend against business risks, ensure platform content security, protect citizens’ image rights and reputation rights, and maintain healthy and pure Internet environment.