German researchers develop a new facial recognition tool that can work in dark
Face detection or recognition is one of the vital technologies that help the law enforcement and security services to catch hold of the criminals.
The new face recognition system that is being used currently would work at its best in case the photos are clean and clicked in proper light. This indicates that when the object has been clicked in poor lighting or in darkness, these technologies would not serve its purpose.
Two scientists from the German university have developed a new face recognition system which works by analyzing the thermal signature of an individual rather than the traditional methods.
Saquib Sarfraz and Rainer Stiefelhagen, these two computer scientists from the Karlsruhe Institute of Technology, Germany, have actually created a new system which first analyzes the mid or far infrared images and then matches these with the standard photographs.
During a regular photo session, the light reflected from the face is captured, which we can see in the form of a standard colored photograph. However, the infrared images which is captured on the basis of thermal signatures emitted by human face is entirely different. Thus matching these two totally non linear entities is definitely a big challenge for the computer system.
In addition, the thermal emissions would also vary depending on the external environment such as weather and over all temperature. It would also vary depending on the skin’s temperature, change in the expression of the individual and the overall physical activity of that particular person.
Also, when an image has been taken by an infrared camera it possesses a lower resolution when compared to a normal photograph. Well, these are some of the most important hurdles which are confronted when one tries to match these two types of images.
The scientists found out a way to overcome this difficulty by making use of a system called as ‘Deep Neural Network’.
The ‘Deep Neural Network System‘ is able to process the pictures and even recognize people in bad light or darkness.
Deep Neural Network System:Â This technology has been designed in such a way that it imitates the function of a human brain. Thus it has the capacity to connect the two different types of images and also recognize people from a a complex set of factors.
The only requirement for the system to give apt results is that it requires a huge database. So scientists had to provide a vast data that comprised of both infrared and visible light images so that the smart system could learn to make the comparisons.
During their study, the duo researchers used the University of Notre Dame set which contained a vast details of both types of images and it also included the images that were shot with different facial expressions and under different lighting conditions. The enormous data also comprised of multiple images of the same person over a certain time period time which helped to provide the photos clicked under different environmental and emotional conditions.
It seems the data used for the study contained around 4,585 of the infrared and visible light images of about 82 individuals.
Further, the scientists divided the entire set into two halves, the two types of images of 41 people were used as “Control” or “System training” purpose and the other set of 41 people was used to “Test” the system.
The experiments proved that the new facial recognition system is much better when compared to the existing facial recognition systems.
MIT Technology Review quotes that Sarfraz and Stiefelhagen say: “The presented approach improves the state-of-the-art by more than 10 percent.”
The beauty of this new face recognition system is that it can match the images and recognize a face in mere 35 milliseconds. The researchers hence say:Â “This is therefore, very fast and capable of running in real-time at about 28 fps.”
The technology has not been rolled out for commercial use yet. Sarfraz says: “There are no plans to roll it out.”
While speaking to BBC, Sarfraz said: “We have been doing research on face recognition already for several years and have a scientific interest in the problem. Our present work on face recognition in thermal images is currently not used outside the research lab.”
Sarfraz feels that currently this technology has a 80% success rate and it has given an accuracy of 55% with one image. Hence he believes that the system requires some more training data as well as much more powerful architecture to give better results.
For now, Dr Tom Heseltine, head of research for UK face recognition company Aurora, seems to be quite impressed with the system. While speaking to BBC he said: “Although the ability to recognize faces in the dark is not new, the ability to use thermal infrared and match against a standard colour photograph could open up some new specific applications areas. Questions may be raised about how susceptible the technology is to variations in body temperature. Their biggest advantage comes in that they could potentially operate in the dark without the need for active infrared illumination.”
On the other hand, Daniel Nesbitt, research director of privacy campaign group Big Brother Watch, told the BBC: “All facial recognition technology has the potential to be very invasive of an individual’s privacy if done the wrong way. This new research makes it even more vital that facial recognition is properly covered by legislation and that strong safeguards are put in place to protect people from misuse. It is important that a serious and wide-ranging debate about all new technology of this type takes place. Our privacy cannot be forgotten no matter how innovative a new piece of technology may seem.”
The researchers, though, feel that the system can be one of the best boons to law enforcement and security officials which will enable them to identity and catch the criminals with some ease.
Readers can check the details of the study which has been published here(PDF).
