How deep fake recognition tools work

11th May 2022
How deep fake recognition tools work

By Praveen Paulose, MD & CEO, Celusion Technologies
May 11, 2022: The digital revolution has transformed the customer onboarding process for enterprises. Since the entire process moved online, the digital verification of customers became critical. The KYC or Know your customer process of verification is conducted online using the Video KYC tool. However, given the number of cyber frauds, it is essential to be vigilant. Deep Fake Recognition tools are used to confirm the authenticity of the identification documents and videos provided by customers. With modern technology advancing at a fast clip, it becomes necessary to separate the genuine from the fake. As technology evolves, it could be misused to tamper with the truth and those in business would need to stay ahead of the learning curve to detect and deal with fakes before they cause damage. Many enterprises wonder if these techniques would pose a real risk to their eKYC processes which could jeopardise their normal customer onboarding.
Let us try and understand how deep fake technology works and how it can be detected.|
What are Deep Fake Videos?|
Deep fake videos are artificially generated videos by combining images to create events or statements that didn't happen. These videos are created using machine learning techniques and AI to generate fraudulent content. Artificial intelligence is used to create fake videos that can easily pass off as real and manipulate the viewer. Digital locations are difficult to distinguish from real locations. Using various technologies, the person in the video can be made to resemble another one. He or she can even be made to mouth words they never used to create a fake video which looks eerily genuine. Deep fakes are a by-product of Deep Learning. Deep Learning algorithms can be used to swap faces in videos to create untrue versions of events. It can be used to create morphed images and audio as well. This technology can be used to perpetrate frauds by impersonating another person to create fake bank accounts or take loans etc. Facial recognition tools need to get smarter to be able to tackle the menace of deep fake videos and images.
Facial recognition
Facial recognition software is used by enterprises to match the face on the identification document with the person in the video while conducting the video KYC. It is important to use deep fake recognition tools to ensure that a fake person doesn’t get through the test.
How do Deep Fake videos work?There are various techniques used to create Deep Fake Videos, the most common being the face-swapping method. Autoencoders are used in deep neural networks to apply a face swap on the target video. They create a series of video clips of the impersonator and then insert them in the target video to replace the person.
Generative Adversarial Networks and their role
GAN is a type of machine learning that helps to create deep fakes that are even more evolved and harder to trace. It works in multiple rounds to improve the flaws in the deep fake and make it tougher to detect by deep fake detectors. With the advancement of technology, the generation of deep fakes is getting easier as even beginners have access to a lot of open-source software.
How do you recognise deep fakes?
Deep fake uses AI to swap a person’s face and voice with that of another to create fake videos that can be dangerous if used by criminals to create fake identities. The technology behind Deep Fakes is the same technology that can be used to identify frauds. Faster algorithms that create deep fakes have become a growing threat to customer identification which can flaw the online customer onboarding process by letting fakes into the system and opening up the possibility of fraud.
Companies can verify users by using a combination of AI and biometric technology but the rise of deep fakes can bypass these biometrics and proof of identity to obtain access to confidential information. Earlier deep fakes were less advanced and could be identified by the frequency of the blinking of eyes. If a person doesn't blink for a very long time or blinks too frequently, it is unnatural and should raise an alarm. However, newer versions of deep fake technology have been able to override this.
A second method of identifying deep fakes is to look at the face, skin and hair which may seem blurred compared to the surroundings. Mismatches in the lighting and the lack of synchronisation of the audio may also be giveaways and indicate a deep fake.
How does it impact Customer KYC
Customer KYC videos embed liveliness detection to confirm that the video is in real-time and the person in the video is live and responding. The sequence of questions is randomly altered to detect a fake subject. Liveliness detection plays a very important role in detecting fakes during video KYC. It prevents fraudsters from using bots, fake videos and forged documents to bypass verification procedures. It is essential to use deep fake recognition tools to weed out fake customers as they can be a potential threat to the system.
Summing up
The creation of deep fake videos is a complex process involving tens of thousands of video, audio and image samples of the person concerned. This would be next to impossible to obtain this quantity of data samples for an ordinary person. Secondly, the video verification process is in real-time and involves the customer displaying the Identification Document. It would be difficult to fake an ID because the absence of a hologram would expose it as fake. The random sequencing of questions is designed to test whether the person is present during the call.
Using a deep fake would not be feasible in a real-time video identification process since the AI would recognise that it is a recorded video. It would be impossible for a recorded video to adjust to the real-time video identification process. AI can identify if the background is a static image and lighting and placement of shadows can also indicate if the subject is fake. These videos are validated by a team of qualified professionals in a backend process which eliminates the risk of identity theft.
While technology is constantly involving using Deep Learning to create higher quality fakes, the methods to recognise deepfakes are also improving. Keeping in mind the presence of deep fake technology, enterprises should be ever vigilant and proactive in the use of deepfake recognition tools to ensure that the customer verification process is smooth and serves its purpose|
 Founded in 2004 and based in Mumbai, Celusion Technologies is a software company that develops Enterprise Applications with cutting-edge technology  for its clientele in the BFSI sector.