New Delhi: The Delhi Police will soon adopt a new software to scan its criminal database using sketched portraits to find the most suitable image matches automatically—much like a reverse image search—to identify suspects in cases with no information available other than an eyewitness version, ThePrint has learnt.
According to senior police officers privy to the details, the buildout is still at a nascent stage. Indraprastha Institute of Information Technology, Delhi, which was awarded the project, is currently developing the software.
Explaining how it would help the police, a senior officer said, “For instance, there is a terror attack. With this software, after preparing the portrait in our in-built system, we can use the software to zero in on suspects, from the pictures in our criminal database. This will help not just in cases where the CCTV images aren’t clear, but also where there is an eyewitness to the crime or someone who has seen the suspect.”
The officer further said that to identify suspects on the basis of CCTV footage, there is the facial recognition system, and in most cases, this is helpful in zeroing in on them.
“However, with this software, the cases where people have only seen the accused/suspects, while leaving the place of the incident or during commission of the offence, we can easily get a grasp of how they look. This is something that the police don’t have yet. Even CCTV footage is blurry most of the time, so that wouldn’t be of much help in face recognition software,” the officer elaborated.
The new software, according to officers, will also work in all blind cases of murder, rape and other crimes, bringing down the time spent on manually comparing the portraits to their digital database of criminals.
Initially, IIIT Delhi had come up with a verbal input-based segregating software. This, officers said, would have selected images or pictures from the database based on verbal and textual inputs.
However, after rounds of iterations, the Delhi Police’s Technology & Project Implementation Division asked the team of engineers to instead develop an interface that can readily filter out the closest images from the database, when provided with the portrait.
“We already have had a portrait-making software for over two decades. Moreover, using verbal or textual inputs would be a lengthier process because everyone’s notions of features are subjective to them. The first software from IIIT that we tested, operated on verbal/textual information, and from the drop-down menu in the dashboard, one could select the categories of facial features, which we could then compare,” the officer quoted above told ThePrint.
In the new project, the model will be trained to work on matching scores. “For instance, a 70 to 80 percent match would be very good. This will help us identify criminals very fast, and also get us the closest looking image to work on. It will also cut down the time wasted in manual search and comparing the portraits with our own system. Moreover, visual identification is a much better form of identification,” a second officer explained.
“The software will work on certain permutations and combinations. They will form clusters and segregate the images. The process will be automated. The model training will make it an interface that will readily filter out, for instance, first the top 15 pictures from the system, and then with commands, it will bring up the top five images,” the officer added.
(Edited by Mannat Chugh)
Also Read: Delhi Police’s newly formed ‘Shishtachar’ squad doesn’t want to be ‘anti-romeo’