DG Research Gives Teeth to FBI Investigations
Speaking to a group of television critics, CSI star William "Gil Grissom" Petersen said that he hears from real-life crime-scene investigators - who tell him that civilians, now aware of the Luminal test for blood traces, demand, "Where's the cool blue light?" Concurred fellow actor Gary Dourdan, who plays Warrick Brown, "When we went to New York, the cops were saying stuff like, "'Everyone that gets his car broken into now wants the whole crime scene done to get his radio back.'"
Unsurprisingly, real life lacks both Hollywood's glamour and its ability to wrap up complicated cases in less than an hour. Take dental identification, for example. If a victim's name is already known, dental records (x-ray images) can be used to make a positive identification from the remains.
But if investigators have nothing more to go on than a piece of jaw, it can take an enormous amount of time to go through a repository of dental records. And while dental matching is such a time-and-labor intensive task, very few identifications have been made using the method alone.
An FBI-appointed task force recommended the creation of an Automated Dental Identification System (ADIS) that would make more effective use of the National Crime Information Center's (NCIC) Missing and Unidentified Persons (MUP) dental files. The goal of ADIS is to computer automate dental record matching in a way similar to what is currently done for finger-print matching. To that end, the Criminal Justice Information Services Division (CJIS) of the FBI has provided West Virginia University with a volume of dental radiographs in support of a Digital Government project to create a prototype system for automating the identification of dental records.
The co-PIs on the project are Hany H. Ammar, a professor in the Lane Department of Computer Science and Electrical Engineering at West Virginia University, and Dr. Anil Jain, University Distinguished Professor in the Department of Computer Science and Engineering at Michigan State University and Dr. Mohamed Abdel-Mottaleb Associate Professor in the Department of Electrical and Computer Engineering, University of Miami.
The team started by discussing their work with CJIS staffers. "We changed some of our initial ideas based on their feedback," says Ammar. "Now we are continuing to modify the system and will eventually go back to them with the prototype."
Building ADIS is a major challenge, involving research in the fields of image processing, computer vision and pattern recognition. The first priority, according to Ammar, is accuracy in matching. "We will first push accuracy, then enhance the matching speed." Image comparison remains one of the greatest challenges of computer science, the one where we most often see the grace and speed of the human brain beat the awkward attempts of silicon. "We have identified several problems that are pushing the state-of-the-art in image processing, image enhancement, and feature extraction," says Ammar.
For just one example, to a human, the concept of "missing teeth" is obvious. We can instantly understand that the three teeth remaining on a section of jaw match three out of six of the teeth in a radiograph taken a decade earlier, while a computer continues to search for a perfect 6:6 match. "The challenge is to look at what is present and ignore what isn't," says the team's forensic dental collaborator, Dr. Robert Howell, Professor Department of Oral and Maxillofacial Pathology, School of Dentistry West Virginia University. If computers may still find abstract concepts challenging, they have a tremendous advantage when dealing with thousands of records. "Humans may be more intuitive, but what about fatigue?" asks Howell.
In the second phase, "The Dental Match Comparison System" the subject's dental records are compared against the candidate records. In this more complicated phase, a subject and a potential match radiograph are put through three preprocessing steps before matching:
After preprocessing, the program then moves to the decision-making stage of the dental match. In this stage, low-level features from the subject and the candidate match are compared to estimate the probability of a match.
|This site is maintained by the Digital Government Research Center at the University of Southern California's Information Sciences Institute.||" CONTACT " POLICIES|