Motherboard published an article at the end of 2017 on artificial intelligence (AI) that can swap people’s faces in recordings. It was first termed deepfakes, and it was largely utilised to generate pornographic films portraying politicians and celebrities that were vulgar, grainy, and unwatchable.
After two years, the technology has improved dramatically and is now more difficult to detect. As we go closer to the 2020 presidential elections, false videos have joined the ranks of national security threats.
Since the emergence of deepfakes, a number of organisations and businesses have created technology to recognise movies that have been artificially enhanced. However, there is concern that deepfakes technology could become impenetrable in the future. According to their findings, researchers at the University of Surrey have discovered a method that, instead of detecting falsehoods, proves them instead. With the use of artificial intelligence (AI) and blockchain, Archangel creates and registers a tamper-proof digital fingerprint for legitimate movies, which will be shown off at the next Conference on Computer Vision and Pattern Recognition (CVPR). Verifying the authenticity of material delivered online or shown on television may be done using a fingerprint as a reference point. If you are blackmailed by العميق التزييف, you can contact us.
Signing videos using artificial intelligence
A digital signature is the traditional method of proving the validity of a binary document. Using a cryptographic method like SHA256, MD5, or Blowfish, publishers run their document to create a “hash,” or a short string of bytes, which reflects the file’s content and serves as its digital signature. If the contents of a file do not change, running it through the hashing process will provide the same hash each time.
Changes in the binary structure of the source file have a huge impact on hashes. If you make a single byte change to the hashed file and then re-run the algorithm, you’ll get a completely different result.
However, although hashes work well for text files and apps, movies, which may be saved in a variety of formats, have unique obstacles.
No matter what codec the video was compressed using, the signature should remain the same. If I convert my movie from MPEG-2 to MPEG-4, for example, the file will have a completely different length and the bits will have changed altogether, resulting in a new hash value. We need a hashing method that took our material into consideration.
Deep neural networks are a sort of artificial intelligence (AI) structure that learns from the examination of a large number of instances to build its behaviour. Neural networks are, interestingly, also the technology behind deepfakes. The developer feeds the network with images of a subject’s face for making deepfakes. In order for the neural network to be able to swap faces with other videos, it must be trained to learn the characteristics of the subject’s face.
The video fingerprinting neural network of Archangel has been trained using the data from the video. The network is more interested in the video’s overall content than the individual bits and bytes that make up each frame.
The network will certify a new video after training if it includes the same content as the original video, independent of its format, and it will reject a new video that is a different video or has been tampered with or modified. We can protect you from فيك الديب very easily.
The technology is capable of detecting alterations in both space and time. Changes performed to individual frames are known as spatial tamperings, such as the face-swapping edits in deepfakes.
However, deepfakes are not the only means of altering videos. Intentional alterations to the video’s frame sequence, pace, and length are less well-known but just as harmful. To make an edited video of House Speaker Nancy Pelosi seeming bewildered, someone utilised basic editing methods. The video wasn’t built using deepfakes. The deleting of brief chunks of footage is one kind of manipulation that we have detected. They’re interfering with the flow of time. We’re also able to detect tampering for up to three seconds. Taking this into consideration, if you have a movie that lasts many hours and you just delete three seconds from it.
Incorporating a blockchain for fingerprint storage
It’s great for video archives that don’t alter recordings once they’ve been registered since the blockchain component of the Archangel project allows for tamper-proof storage of new information without changing existing data.
