Is that an elephant running towards you or a pile of rocks? Since your life may depend on it, your brain makes a decision in less than 20 milliseconds.
A new software model of the visual cortex may explain how we do it. It could also point to a new way to build vision software for robots and surveillance systems.
We can tell if something is an animal or a face before we are aware we have seen anything at all. One idea is that an initial 鈥渟weep鈥� of neuronal activity may accomplish this before feedback loops that refine our perceptions kick in. So Tomaso Poggio and colleagues at the Massachusetts Institute of Technology have built a computer model that mimics the groups of neurons in the visual cortex and the way in which they process information. One group will just look for edges, for example, and another for corners. The model passes data from one neuron group to the next in the same way as the brain does, starting with basic image-processing neurons and moving up to those responsible for more sophisticated recognition. Unlike the brain, however, it cannot feed data higher up the chain back to an earlier neuron group for further refinement.
Advertisement
Nonetheless, when the model was 鈥渟hown鈥� 150 animal images and 150 non-animal 鈥渄istractors鈥�, it picked out the animals as well as people do when shown the images for just 20 milliseconds (Proceedings of the National Academy of Sciences, DOI: 10.1073/pnas.0700622104). This suggests that we may bypass our feedback loops in rapid recognition.
