Usually, victims or witnesses can’t do more than describe the suspect. They don’t have time to take pictures and/or security cameras are not around. Words are all forensics specialists rely on when coming up with a sketch. From that point, it may take a while until the suspect is found, period that would be significantly reduced if there were pictures to go around. Scientists at Radboud University thought the same thing when they developed an algorithm that can turn mere drawings in photorealistic faces #machinemagic
With the help of deep neural networks, scientists produced pictures of the people drawn, a process that could be very useful in investigations but also various art forms. This algorithm would simplify police officers’ lives since they could simply apply face recognition tools to suspect photos, not sketches.
In order to teach the machine to turn sketch to photo, the scientists had to come up with a dataset that had sketches and photos that matched. The machine then received only the drawings and started turning them in images. At first, the machine would not make images identical to those in the dataset but as mistakes were pointed out, it started to learn a winning strategy. Everything happens gradually in machine learning, anyway.
The method is explained in details here , including the information used to help the machine learn. We’re excited to see how will the technology be used at its full potential.
Follow TechTheLead on Google News to get the news first.
