Facial recognition: Coming to a watch near you soon?
You can already control your mobile devices with touch, gesture, or voice commands. Now Finnish researcher Olli Lahdenoja is introducing another mechanism: facial recognition.
Lahdenoja, a researcher at the University of Turku, has published a doctoral dissertation which develops a facial recognition method that can fit onto a single electronic chip. The system has the potential to be used in devices such as smartwatches, as well as mobile handsets, thanks to its compact size and low power consumption.
"[Using this system] a smartwatch could, for example, switch on when a user looks at it. What's more, the recognition system could be used to identify a person when they access an online service," says Lahdenoja. "The actual image sensor and the circuit where computations are carried out are embedded into the same chip. Its physical size is small, and the only thing required is a power source."
While other facial recognition systems already exist, Lahdenoja isn't aware of any others that fit the whole system onto a single chip.
Lahdenoja's system is based on the Local Binary Pattern (LBP) method, a widely-used facial image analysis method also developed in Finland. What Lahdenoja has done is apply LBP to focal-plane vision processors which combine photodiodes with processor circuitry to enable more efficient computation and high-speed image analysis.
Although LBP dates back to 1990s, until now it hadn't been applied to focal-plane processors. The LBP method measures facial texture which is derived from grayscale values in an image.
"A pixel is taken from an image and compared to its neighbouring pixels [to determine] whether its grayscale value is higher or lower. These values are used to create binary chains," Lahdenoja says. "With LBP this can be done to an area in an image. All pixels from the area are taken and compared with its neighbours. Histogrammes are then created based on how many binary chains there are in that area."
In the actual facial recognition element of the method, an image of the face is divided into smaller areas. Histogrammes from all these areas are then compared with two images of the same face. The more the histogrammes match, the higher the possibility it is the same face. Lahdenoja's method also recognises a face when it is moving.
It may not sound like a fast or precise method, but according to Lahdenoja typically around 97 percent accuracy is reached depending on the quality of the image.
"The benefit of focal-plane processors is that every pixel is surrounded by calculation logic which means calculations can be done very fast. We can talk about speeds of up to 100,000 images per second," says Lahdenoja, but clarifies these speeds cannot, at least not yet, be reached with facial recognition.
Goodbye passwords, hello facial recognition
While the new face recognition system has many potential practical applications, such as identity verification for shopping online or social media, it's likely to be some time before it becomes publicly available.
Lahdenoja doesn't see any major issues in applying his methods in practice, but so far the chip remains on a concept level and has only been partly tested using existing focal-plane processors. The actual chip building would require a commercial partner.
Nevertheless, we could be saying goodbye to passwords sooner than many think. Earlier this month Tsinghua University in Beijing announced it is developing an ATM with facial recognition. Furthermore in March, Chinese e-commerce giant Alibaba demonstrated software for buying goods using your face.
Alibaba also predicted facial recognition will be the next big thing in online payments: Lahdenoja's could be about to come in handy.
Read more