February 2, 2015
You may worry that intelligent robots will replace humans any day, but that isn’t happening anytime soon. For now, the best computer algorithms cannot do even simple visual tasks like recognizing distorted letters. This is exploited each time we are asked to recognize distorted letters on website. These tests – called CAPTCHAS (for Completely Automated Public Turing test to tell Computers and Humans Apart) – are ubiquitous on the internet because they can prevent access to malicious computer programs. So what makes our brain so good at vision?
Through decades of research, neuroscientists have now found that there’s much more to vision than meets the eye. The eye works much like a camera. Light enters through the pupil and the lens focuses light onto a screen called the “retina”, which is akin to a camera film. Neurons leaving the retina carry information about the image to the visual areas in the brain, which occupy as much as 40% of the total real estate in the brain. This disproportionate area occupied by vision in the brain shows that vision is not easy for the brain either.
Dr SP Arun and his team have been studying biological vision at the Centre for Neuroscience, Indian Institute of Science, Bangalore. In a recent study, Arun and PhD student Ratan Murty have shed light on how the brain interprets the 2-dimensional image falling on the retina. “The image on the retina contains relevant as well as irrelevant information,” Arun says, “The same object can produce different images because of changes in lighting, size, position and three dimensional rotations. These irrelevant variations have to be factored out by the brain for it to understand that all these images belong to the same object. This computation is performed by neurons in the visual cortex.”
Ratan and Arun have performed recordings from the inferior temporal cortex of the monkey brain — an area that is known to be crucial for visual object recognition. They have found that flashing an image results in neural activity that builds up and drops over a period of time. During the build-up of the response, neurons are sensitive to irrelevant variations such as changes in the view point of an object. But in the later portion of the response, neurons respond to the same object ignoring irrelevant stimuli. “This transition from view dependence to view invariance has never been shown before, and it shows that neurons in this area perform this important computation dynamically over time”, said Ratan.
Ratan and Arun are performing a number of other experiments to understand how the brain processes three dimensional information. “Precisely how the brain ignores all the irrelevant variations is a fundamental problem in vision,” Ratan adds, “My experiments will help us understand at least the problem of viewpoint invariance better.” The researchers believe it is something that the brain has learned to solve over the course of evolution. Robots may beat us at algorithmic games like chess but they are nowhere near human competence in real-world tasks like vision.
The paper appeared online in the Journal of Neurophysiology during early 2015. http://jn.physiology.org/content/early/2015/01/16/jn.00810.2014