javascript contador Skip to content

From robot portrait to photograph of the suspect with the "magic" of artificial intelligence

From robot portrait to photograph of the suspect with the "magic" of artificial intelligence

From robot portrait to photograph of the suspect with the "magic" of artificial intelligence

The use of artificial intelligence It is not only useful to order pizzas through Facebook Messenger, it would also be of great help to identify criminals.

Artificial intelligence can be applied to hundreds of different fields, a few days ago we looked at an algorithm that used machine learning to color images in black and white without knowing the original colors at all.

When it comes to making a sketch of a person’s faceWe humans can do it with more or more precision (it also depends on drawing skills, of course), but if we resort to computers there are dozens of programs and applications that do this task in an instant.

From sketches to realistic portraits thanks to artificial intelligence

fearartificial intelligence11

And on the contrary?Could we create an artificial intelligence that recreates a face with total fidelity having only a few lines of its sketch? The revolution of robot portraits is underway.

From Radboud University in the Netherlands, they have developed an AI using deep neural networks trained to convert sketches and portraits of faces with a few lines to images that recreate with great precision the original face, as if it were a photograph.

What use could something like this have? Well, on the one hand it could be used in the world of art and design, but it would also have a great role if a reliable product could be obtained that could be used by the authorities, for example in the forensic science or as a recognition system in research with basic robot portraits.


Deep neural networks for CSI: Convolutional Sketch Inversion

For train artificial intelligence, Yagmur Gucluturk and Umut Guclu (those responsible for the project) used the information from thousands of images of the Network, using over 200,000 CelebA celebrity faces, in addition to 13,000 from LFW (Labeled Faces in the Wild). The process was to first convert those photos to sketches, repeating the operation over and over again.

After training, the AI ​​was able to go from the sketches to an image very very similar to the original photograph, in some cases even recovering hair color and contrasts when there was no information in the sketches. Of course it is not a perfect system and it still has to improve, but they are sure that by learning the neural networks they will be able to recreate photographs of faces from sketches never seen by AI.

Much remains to be investigated, but they have already created a company called Neurant that is developing this technology, which they have named CSI: Convolutional Sketch Inversion.