Deepfake Technology
The present-day digital technologies make it harder and harder for the human brain capacity to detect what are the real technologies and what are fake. Deepfake technology is one of the most recently developed and emerging technology which marked a turning point in the creation of fake content.
What is Deepfake Technology?
Deepfake technology is powered by the most advanced technological methodologies of Artificial intelligence and machine learning. The main task of this technology is, generating hyper-realistic fake content based on artificial intelligence such as videos, pictures, and audio clips containing the things someone never said or did. The worst effect of this deepfake technology is the possibility of deceiving a targeted person or entire society is endless because of the hardness of detection. With the latest social media technologies, deepfake content can be spread among millions of people within few seconds which will implement a negative impact on society undoubtedly.
World’s first Deepfake
The basic purpose of deepfake technology, which is to swap a real fake with another fake face was firstly raised around 1865. It was the first known attempt to change the face of someone with another face using a painting. In this deepfake former US president, Abraham Lincoln’s face was swapped with the body of John Calhoun who was a politician in the southern US. After Abraham Lincoln’s death, demand for that deepfake painting was surprisingly increased. In the modern world, photo manipulation technology was first introduced in the 19th century, and the same kind of technologies developed later to manipulate videos as well as images. During the 20th century, deepfake technologies improved along with the development of digital media.
Deepfake creation process
The first strive of creating deepfake content was the module named Faceswap which was developed utilizing the technology called autoencoder-decoder pairing structure. Autoencoder is used to capture the features of the face and auto decoder is used to reconstruct that face image. To swap the source person’s face and targeted person’s face, two more encoder-decoder pair is needed. All the parameters of encoders are shared among two pairs of encoders which enables to find of the similarity of the two faces. This process is relatively simple since the basic shapes and properties are the same in every face such as the position of the eyes, mouth, and nose.
Faceswap is the first known tool that is capable of creating deepfake content. When the face swap application was initially developed and released, that technology was unconventional. It has been an enormous step in artificial intelligence developments because the source code of this application was completely opposite of the current theories and it was confusing.
Models such as Faceswap-GAN, Few-Shot Face Translation, DeepfaceLab, and Dfaker can be also considered as examples for deepfake creating tools.
Deepfake detection process
Usually, deepfake detection tools using a binary classification system to examine and classify real content and fake content. This kind of strategy requires a database containing a huge number of real and fake videos to program the classification module. Dmitry Korshunov and Sébastien Marcel were able to create an outstanding deepfake dataset comprising 620 videos generated based on the GAN model utilizing an open-source code named FaceSwapGAN. VidMIT database which provides audio and video clips freely were used to create both low and high-quality deepfakes. Those videos were utilized to test different deepfake identification techniques. Test outcomes show that famous face recognition frameworks dependent on VGG and facenet are unable to detect deepfakes effectively.
Apart from that methodologies, deepfake videos can be detected by some face properties as well. Eye blinking is one of the most prominent properties that helps to find deepfake content. Most of the face images available in the datasets are images of faces with opened eyes. Therefore, deepfake video creating tools are not having the capability to create fake videos with faces with blinking eyes that are blinking in the normal frequency.
Uses of Deepfake
Deepfake technology can be used successfully in Blackmailing. In a real blackmailing scenario, someone requesting some advantage from another one is a trade-off for not revealing sensitive information about them. But using deepfake technology, someone can easily create fake content about a targeted person and request advantages for not revealing it to the public. This can be detrimental for both parties since even if someone stole some sensitive content of another one and try to blackmail them, the victim party can say it is fake content made using deepfake technology.
The film industry can be considered as an example where use deepfake technology for good purposes. For example, Movie directors can create a younger version of an actor for the prequel movies that have been become very popular recently. On the other hand, movie producers can use deepfake technology for the actors who passed away during the production of that movie. Placing Harrison Ford’s younger version as young Han Solo in “Solo: A Star Wars Story” is one example of that incident.
As described under that points, obviously deepfake technology is not an area that will wind up being developed. The nonstop evolvement of the deepfake content-making strategies just as deepfake recognizing strategies is the main reason for that. If we can raise the usage of deepfake technology for good purposes, there is no doubt that there will be a revolution in digital media technology in the near future.
Thank You
References — https://en.wikipedia.org , https://github.com , https://conradsanderson.id.au/vidtimit , https://scholar.google.com
