A Visionary Approach to Artificial Intelligence

We often forget how complex human vision is... and that poses plenty of challenges for those trying to teach machines how to "see".

 |  Transcript [PDF]
Share

“One thing that’s really exciting about studying vision is that it is so powerful that we don’t actually think about it from day to day. We just use it. It’s only when we pause and think, ‘wait, how do I actually do this?’, that we start to think about just how complex the problem is.”

Vision scientist James Elder, professor of psychology and electrical engineering and computer science at York University, is a member of York’s Vision: Science to Applications (VISTA) program. His multi-disciplinary team is working to better understand human vision and perception in order to train more intelligent artificial vision systems. He calls this biologically-inspired computer vision, building on the powerful and adaptive human visual system to teach computers how to interpret what they “see” in video input.

While human vision feels simple and automatic, the artificial intelligence (AI) community faces many challenges in replicating such a complex processing system in computers. For one, people perceive the world as being 3D, even with one eye closed. This remains something that machine vision struggles with, says Elder.

Another feature of human vision is that people can easily attend to a specific part of the scene to focus on and process. Machine vision still lacks that ability to pick out and be attentive to a narrow part of its field of view.

A third aspect that machines struggle with is being adaptive. For instance, new environments, or even changes in weather, can be enough to throw off machine vision. Human vision can readily adapt to the new context of changing locations.

Human vision is still so much more powerful than machine vision. Studying vision systems in this way gives a greater appreciation for just how complex human vision is, and how remarkable it is that these feats all feel effortless.

“Vision in particular is a really cool thing to study because we experience it every moment of our lives,” says Elder. “We can do this real diversity of tasks: recognizing people, appreciating beauty. It’s kind of automatic, and yet there are, in secret, these billions of neurons working away.”

‹ Previous post
Next post ›

James Elder is a professor and York Research Chair in Human and Computer Vision at York University. He is jointly appointed to the Department of Psychology and the Department of Electrical Engineering & Computer Science at York, and is a member of York’s Centre for Vision Research (CVR) and Vision: Science to Applications (VISTA) program. He is also Director of the NSERC CREATE Training Program in Data Analytics & Visualization (NSERC CREATE DAV) and Principal Investigator of the Intelligent Systems for Sustainable Urban Mobility (ISSUM) project. Elder's research seeks to improve machine vision systems through a better understanding of visual processing in biological systems. His current research is focused on natural scene statistics, perceptual organization, contour processing, shape perception, single-view 3D reconstruction, attentive vision systems and machine vision systems for dynamic 3D urban awareness.

Research2Reality is a groundbreaking initiative that shines a spotlight on world-class scientists engaged in innovative and leading edge research in Canada. Our video series is continually updated to celebrate the success of researchers who are establishing the new frontiers of science and to share the impact of their discoveries with the public.