Emerging for the first time in 2018, deepfakes They became known for making it possible to swap the faces of actors from one video to another. The word Deep comes from the term deeplearning (deep learning) a branch of AI that involves algorithms that accurately replicate the learning capabilities of humans and animals.
There is a type of neural network used in forgeries called autoencoder (autoencoder), which is responsible for encoding an input image in a small set of numerical values with the help of layers.
The initial layers with many variables are simplified until reaching the bottleneck layer, which is then decoded by the neural network to recreate the original image.
A series of images serve as a source of data to feed the autocoder during its training process so that it can find a way to adjust the parameters corresponding to the encoding and decoding layer until the output image is as accurate as possible. to the entrance.
During training, the autocoder is supplied with a series of images. The goal of training is to find a way to adjust the encoder and decoder layer parameters so that the output image is as close to the input image as possible.
And although this technology is still not trivial, deepfakes have made video manipulation a resource available to everyone, thus causing the appearance of forgeries focused on the generation of fake news.
This has led AI researchers to take the initiative to develop tools that allow deepfake detection. One of them was developed with the purpose of marking those videos where the person did not blink or did so at abnormal intervals.
There is another method that contemplates the use of deeplearning algorithms designed to detect signs of manipulation at the edges of objects in images.