The clothes may make the man, but if a University at Buffalo professor has his way, the shoes will nab the criminal.
Dr. Sargur Srihari, a computer science professor, is building a search engine populated with thousands of shoe images scraped from internet shoe stores that would let police forensics units submit a photo of a shoe print from a crime scene and quickly learn the gender, size and brand of shoe a killer or thief was likely wearing.
Shoe prints are some of the most common evidence left at crime scenes, but there's a limited number of shoe identification experts and the tech tools available to them aren't nearly as good as one might think from watching cops shows and movies. Complicating matters, shoe treads wear down and wear out, and small pebbles lodge in a shoe's ridges, distorting their prints. That's where computational forensics comes in.
"CSI makes it looks like it is all solved," Srihari said. "But actually, it is a very hard problem. Shoe prints can even harder than fingerprints to match."
Srihari and his grad students started by scraping online shoe stores like Zappos for images of the bottoms of shoes. Then they analyzed the images to identify patterns and built a shoe tread database of more than 10,000 pairs of men's shoes, ranging from Timberland boots to Nike indoor soccer shoes.
Srihari was surprised to find that many online web sellers include pictures of soles of shoes.
"It just amazes me there are so many unique patterns," Srihari said. "People like individualistic shoes and I guess the look of the sole matters to people."
Then to simulate crime scene evidence, they had students step in powder and walk on a carpet to create a "positive" images and step onto a dusty surface to capture "negative" images.
Now, he's working on algorithms that will identify likely matches for his images and for partial foot prints. Working from the suggested results, a forensic shoe print examiner can find the most likely match, which can then be used to build a criminal case.
Srihari's approach won $300,000 in grant money this year from the Justice Department's National Institute of Justice, which funds research into new technology and approaches to law enforcement. Currently shoe print examiners rely on deep human knowledge, intuition, and an online system for asking for help from the small handful of people who know a shoe by its print.
"Shoe print forensics is not such a glamorous field," Srihari said. "What we do is take the areas where human intuition comes in and try to use computational tools."
As an example of the power of such techniques, he points to work he did with the U.S. Postal Service in the 1990s when nearly all hand-addressed envelopes were hand-routed. But pattern matching technology largely solved that problem -- now, 95 percent of hand-addressed mail is read only by computers, according to Srihari.
Srihari doesn't expect that kind of success any time soon for the shoe project, which he describes as one of the toughest he's tackled, mostly due to the wear and tear that shoes endure.
Still Srihari says any would-be criminals would be smart to avoid sneakers.
"Go in a suit if you commit a crime -- there are no prints on dress shoes," Srihari said.
No comments:
Post a Comment