Visual Strategies for Complex Tasks
In my lab, we aim to discover the neural strategies that enable human observers to accomplish complex perceptual tasks, and to use these strategies towards designing intelligent artificial systems.
In one project, we are working towards explaining 3-D shape percepts with a detailed model of neural responses in the cortex, based on the original discovery that 3-D shape percepts from texture cues are automatically linked to patterns of orientation flows in retinal images whereas spatial frequency gradients are used to infer relative depth. Neurons at the first stage of the visual cortex serve to parse the retinal image into orientations and spatial frequencies, so computational, psychophysical and electrophysiological techniques are being used to decipher how these early responses are combined by later neurons to extract the critical orientation flows. These studies have shown how to explain variations in visual percepts by variations within neural populations, and how early processes in the cortex enhance the processing of critical information by later stages, but the really knotty issues have yet to be tackled.
Our recent work in color perception has concentrated on neural processes that categorize and identify objects across different illuminations, despite the absence of shape or texture cues. We have shown that color signals from everyday objects change in a collective manner across natural illuminants, and though this would make it easy to discount the illuminant, it is easy to design algorithms that extract the shift and give both relative colors of objects and information about illumination changes. These algorithms were used to design identification experiments on real objects and lights. The results have shown that human observers use heuristic strategies that are simpler than these optimal algorithms but not as accurate, as if they consider the costs of computation in solving complex tasks. These strategies have raised new kinds of questions about the use of color information by observers and the geometry of perceptual color space, which are being pursued by graduate students and post-docs in the lab.