When you go to polling booths to cast your votes this Thursday (April 27), you may feel intimidated by digital cameras that capture snapshots of your face. Using facial-recognition software, officials at the Election Commission (EC) of India are trying to match your face with those of hundreds of thousands of registered voters in their database. The technology is meant to weed out people who are trying to vote twice, says Debasish Sen, chief electoral officer of West Bengal. The use of face-recognition technology for voter identification and authentication was introduced in India when Bihar went to the polls last year. This relatively inexpensive technology has helped the EC eliminate fraudulent voters who show up under several aliases ? and deprive genuine people from casting their votes ? to sway the election in favour of a particular party. The software has also been successfully used in the US, Mexico, Uganda and several other countries to do away with bogus voters.

Non-intrusive

Face recognition is ideally suited for a voting system, provided a proper database of genuine voters exists, says Dr Malay Kumar Kundu, professor in charge, machine intelligence unit of the Indian Statistical Institute in Calcutta. It automates a very complicated process of identification in a split second. According to Kundu, facial recognition is one of the hottest areas among the larger group of technologies known as biometrics, which uses biological information to verify identities. The basic idea behind biometrics is that our bodies contain unique properties that can be used to distinguish us from others. Besides facial recognition, traditional biometric authentication methods have used fingerprints, palm prints and voice identification.

The advantage of face recognition technology is that it is non-intrusive, which means it can be used clandestinely even if an individual doesn’t co-operate, says Kundu. In other words, the technology can even be used to secretly pick up someone’s face in the crowd for surveillance. Which is why the technology is also being used to pin down criminal and suspected terrorists in crowded places, he adds.

People have an amazing ability to recognise and remember thousands of faces, says Dr C.A. Murthy, head, machine intelligence unit, at the ISI. Murthy is working on a project on face recognition at the institute. Computers are being used to mimic that innate ability by turning a face into a digital code so that it can be compared to thousands, if not millions, of other faces, he adds.

If you look in the mirror, you will see that your face has certain distinguishable features. The peaks and valleys ? such as the tip of the nose, the depth of the eye sockets and the distance of the jawline are known as control points or nodal points on a human face, says Murthy. While a human face has 80 nodal points, only 14 to 22 are used for recognition. We concentrate on the inner region of the face, which runs from temple to temple and just over the lip, called the face blob or golden triangle, says Kundu. According to him, a good caricaturist has an eye for the nodal points which inspires him or her to amplify, say, Atal Bihari Vajpayee’s lips or Tony Blair’s ears.

Mapping contours

The heart of the system is a mathematical technique or an algorithm called Local Feature Analysis (LFA) or Principal Component Analysis (PCA) which maps the contours and creates a faceprint or a unique numerical code of a particular face. Once the system has stored a faceprint, it can compare it with the millions of faceprints stored in a database. Under optimal conditions, the software should be able to match a new face with a database in four out of five cases, says Kundu. However, a lot depends on the clarity of both the photos in the database and the images being captured and searched. The match rates could fall if the face is recorded at an odd angle or in dim light.

3D images best

The system is not foolproof, however, says Murthy. Most of these software programs innovated in the US or Europe do not account for faces with long beards or turbans on the head, he says. Incidentally, the software used by the EC was turned out in France. Facial hair or wrinkles on the face could also upset all measurements in the algorithm, he adds. In addition, the software works best when the image is three-dimensional and is captured through a video camera from different angles. If you capture a mug shot through a simple digital camera, what you get is two-dimensional image. If it’s a 2D and black and white image you lose a lot of vital information such as the skin tone. However, face-recognition software launched by the US-based company Identrix claims to have eliminated all these problems and has shown nearly 100 per cent accuracy.

It remains to be seen if the cutting-edge technology used by the EC makes the polls free and fair in West Bengal.