By Nicole Stevenson
Researchers at Michigan State University have developed software with the capability to match police sketches to mug shots, meaning increased efficiency in apprehending dangerous criminals. Previous attempts at incorporating technology into the sketch process focused on using computer software to create digital images of suspects, but until now all have failed work as well as old-fashioned forensic artists. With this new software development the sketch can now be matched to an actual photo of the criminal. This would be a great addition in the field of forensic science and law enforcement.
Michigan State doctoral student Brendan Klare worked closely with Computer Science and Engineering professor Anil Jain to create the software, which was then tested using mug shots of various suspects. Each mug shot had its own matching forensic sketch in the database. The results were impressive: “Using a database of more than 10,000 mug shot photos, 45 percent of the time we had the correct person,” Klare says. Jain explains that the system works by matching “high-level features from both the sketch and the photo; features such as the structural distribution and the shape of the eyes, nose and chin.” While the 45 percent accuracy rate is a great improvement over current facial recognition software, there are still cases in which the new program selects the wrong mug shot based on a given sketch. For this reason, the software is not meant to stand alone as a crime fighting instrument. Rather, the program is meant to speed the process of tracking down suspects—it is a tool to be used in concert with other evidence and questioning by officials.
The implications of such advanced facial recognition technology are huge. Police sketches of criminals have long been used in locating suspects, and while these visual aids are particularly useful in disseminating warnings to the public, they are far from perfect. These renderings have the dual ability to alert communities while at the same time petitioning them for additional information about crimes. However, because professional artists must draw suspects based on verbal description, the witness must convey the image accurately in words. This is a difficult task, and one that can sometimes produce misleading sketches. Forensic art is not an exact science; criminals can change their appearance, details may be rendered inaccurately, or witness memory could simply be faulty. That being said, the police sketch is still the best way of creating a visual representation of a witness’s memory. Now, with the introduction of Jain and Klare’s software program, law enforcement will be able to more efficiently match witness memory to concrete evidence.