Artificial intelligence and display measurements seem like an odd pair. But one of the many ways in which researchers are improving the capability of AI is coming directly from image recognition, and one of the ways to improve images is through strict photometric standards. A new technique using colorization from a team at the University of California, Berkeley, has shown just how true this can be.
By using clumps of data, artificial intelligence in coordination with users can color in gray-scale images. This is achieved through a system called Convolutional Neural Network (CNN). The system directs users based on subtle cues gathered through mounds of data, and additionally, without the direct help of users, the AI can fill in some colors on it’s on.
The result is a steadily growing technology whose base of knowledge continues to expand. CNN has the ability to learn which colors relate to which common images, and can promote these colors in future data sets.
Automatic colorization systems are not entirely revolutionary; they’ve been around for some time, in fact. What sets the CNN system developed at Berkeley apart from its predecessors is the ability to function in real-time and to assist users in grayscale colorization with eerie accuracy. Another area where the CNN system shows improvement was an understanding of when a user was wishing to use a color outside of the normal parameters of a common data point (painting grass purple, for example). It is, in all reality, a next generation AI technology.
The researchers at the University of California, Berkeley, tested this new technology thoroughly and with positive results. The standards they placed on the AI were so stringent, that neophyte users working with CNN could produce astounding results. Famous images rendered in greyscale were filled in by amateur users with the help of CNN to nearly replicate the original representation.
This is quite a leap forward for AI image recognition, and undoubtedly opens paths for new ways to deploy the technology. It’s also a novel way of using display measurement to the furthest limits. If you’re interested, check out our OL 770-DMS and Aries spectroradiometers and our Mantis imaging colorimeter to see for yourself.