Columbia University and Neuromatters LLC announced that they have entered into an agreement to develop a novel brain-computer interface technology for rapid identification of relevant images.
The human brain reacts to images of interest at a pace that is far faster than a person can consciously register. Researchers at Columbia University have developed a technology, Cortically Coupled Computer Vision (C3Vision), that takes advantage of this near-subconscious ability and pairs it with the processing power and efficiency of computers for rapid identification of images that the brain finds relevant.
C3Vision relies on a suite of patented machine learning algorithms which are trained to recognize what is of interest to a human viewer in a given context. Wearing an electroencephalography cap with electrodes that capture electrical activity of the brain, a person is shown a sequence of images at a rapid pace. Each time an image of interest is displayed, the brain emits a distinctive electrical signal which is captured and decoded by the technology. Based on the strength of these neural responses, images are ranked. Over time, the technology learns what types of images are of interest to the viewer, and can eventually identify such images on its own.
"Computer vision systems are good at crunching through lots of data, but rather poor at characterizing images and scenes based on abstract and subjective concepts such as 'that's interesting' or 'I like that,'" says Paul Sajda, Director of the Laboratory for Intelligent Imaging and Neural Computing and a Professor of Biomedical Engineering and of Radiology at Columbia University's School of Engineering and Applied Science and Columbia University Medical Center, respectively. "In contrast, the human visual system is exceedingly good at abstract and subjective scene analysis and image understanding, though would obviously be overwhelmed by having to analyze information from millions of images."
With a combined $4.6 million of support from DARPA, Neuromatters and Columbia are collaborating on the development of an integrated image triage system based on the C3Vision technology. The system will be used and evaluated in operational environments by government image analysts to examine vast areas of satellite imagery for specific physical characteristics. The system may also extend to video surveillance and security, where the aim is to identify suspicious activity.
"We are very excited about our collaboration with Columbia University on the DARPA program and look forward to exploring new applications for C3Vision," says Barbara Hanna, CEO of Neuromatters. "Our goal is to push the envelope of brain-machine interfaces and establish them in areas where information overload is already emerging as a huge problem, and C3Vision has the potential to provide transformative possibilities to help address this problem."
Donna See, the officer at Columbia Technology Ventures who manages the C3Vision patent portfolio, believes there could be significant potential for the technology in consumer markets. "We can certainly envision C3Vision applied to image-rich online contexts, such as fashion, furnishings, even real estate and travel, learning the preferences of tastemakers and buyers to create truly customized and targeted retail experiences."