The Victorian Institute of Forensic Medicine (VIFM) and Monash University are using machine learning to reconstruct faces and identify the dead, following calls for more empirical forensic evidence to replace older methods.
The technology could be used to both assist prosecutors in court cases, and identify the dead following mass disasters.
In an interview with Innovation Intelligence, Associate Professor Richard Bassed from Monash University and VIFM discussed the project.
“This whole process is part of increasing the validity of empirical evidence for the courts.
“What does a face look like on a burnt skull? At the moment that is not accurate quite a lot of the time.
“What we are trying to do is to develop machine learning algorithms using our database as a training set, that can detect and look at a skull and then be able to make the face of that skull based on all the other faces of skulls that it has in its database.”
The VIFM database of post-mortem computed tomography (PM-CT) includes 75,000-85,000 full-body CT scans, and multiple high-resolution optical photographs of the dead.
This PM-CT database sits alongside a case management system (iCMS) that can currently be searched for keyword causes of death.
The researchers also want to discover if the AI system could match the faces of the deceased to photos that existed while they were alive.
According to Bassed, the team is asking “Would it do the same if you were dead and decomposed? Would it do the same if you were dead and had a hole in the middle of your face?
“The application for the project would be mass disaster events. We would take lots and lots of photos of deceased people, put them through a machine, and with passports, licenses and identifiable photographs, we’d theoretically be able to match people up using facial recognition software.”
Bassed notes the database is constantly growing, and includes scans of faces from people of different ages, sexes and ethnicities.
“The CT scan shows you hard tissue, the skeleton, and it also shows you soft tissue: the skin, fat, and muscle that lies over the skeleton. It also reveals the relationship between the soft tissue and the hard tissue.”
Before this technology, people used to stick pins into dead peoples’ faces to identify where the muscle, fat and skin was, and then approximate the face based on the average tissue depths of a skull from their results.
“Usually there’s always been artistic license in the practice, this project is about trying to put some science on top of that.”
Bassed also explains that it won’t be taking away jobs from forensic artists but will allow them to create images with more accuracy.
“This project will produce a face shape that can then be made more realistic by the application of forensic artistry to provide skin colour, hair, etc. Forensic artistry will still have a place, but we hope it will be more accurate.”
Bassed estimates the project will take two to three years before they will have their results. Many participants of the study are University students that are interested in machine learning, with teachers volunteering to supervise.