DepthFake, a new method to detect deepfakes using artificial intelligence
The arrival of the deepfakes generated divided reactions: on the one hand, a positive astonishment was manifested by the remarkable quality of false videos and images that can be generated; while on the other hand, it also raised concerns about its potential use in fraud and other questionable ethical practices.
This phenomenon posed a new challenge for investigations in the area, mainly those focused on security issues. Along these lines, a recent project proposed a method to strengthen the AI-based detection mechanisms used to identify false images or videos.
An additional factor to analyze during the search for deepfakes
A research team from the University of Rome La Sapienza, Italy, proposed a new method to detect digitally manipulated faces, through the identification of depth inconsistencies.
The RGB and depth characteristics, exemplified in the image that accompanies this note, contain semantic information that is easier to interpret and robust to strong compression operations.
This exercise can be done manually with image editors that have advanced tools, such as Photoshop or Gimp. However, the detection of these irregularities may fall to artificial intelligence, since automatic learning systems have already been developed to carry out this task.
The La Sapienza team presented DepthFake, a proposed new method that integrates the aforementioned depth analysis. The peculiarity of this approach is its ability to extract the depth of a face with a monocular estimation method, which is concatenated to the RGB image. In simple words, the decomposition of colors that is shown as an example to detect irregularities, is reduced to grayscale to speed up the search, taking advantage of elements of the image that are not lost with the conversion.
The Article which exposes the main technical aspects of this method and the good results obtained during the tests of its application, is positive evidence of the reactive nature of the development of security tools.
By having such a powerful tool in front of us, these efforts are an approximation to the guarantee of being able to access more interesting audiovisual experiences than fraudulent and malicious content.
