We've seen it in numerous TV shows and movies – the witness to a crime looks through a book of mug shots, then works with a police sketch artist to come up with a likeness of the nasty person they saw. After looking through hundreds of mug shots, however, it's possible that the tired-brained witness could look right at a photo of the guilty party and not recognize them. It's also possible that there is a mug shot of the criminal on a database somewhere out there, but that this particular witness will never see it. A computer system being pioneered at Michigan State University, however, could be the solution to such problems – it automatically matches faces in police sketches to mug shots.
The MSU system doesn't work on a pixel-by-pixel basis, but instead uses algorithms to match faces based on prominent features such as the structural distribution and shape of the eyes, nose and chin. Given that sketches sometimes end up not looking all that much like the real person overall, but retaining some of their underlying characteristics, it becomes easy to see how a computer program could make connections that people looking at a "WANTED" poster might miss.
The system was put to the test using a collection of sketches from crimes in which the criminal was later identified, and a database of over 10,000 mug shots. It was able to match 45 percent of the sketches to the right photos – at this point, perhaps not good enough to base a conviction on, but certainly an aid in guiding investigators to likely suspects.
A field test of the system is planned to take place in about a year.
The research is being published in the journal IEEE Transactions on Pattern Analysis and Machine Intelligence.