Vision scientists have made significant progress in understanding how humans detect a single stimulus, but the vision community knows much less about how we process complex scenes consisting of multiple stimuli. The issue of how we combine information from individual stimuli to get the big picture continues to be an important, albeit difficult problem. At present, there is considerable evidence that the visual system decomposes the visual scene using local mechanisms tuned for specific stimulus properties such as spatial frequency, orientation, direction of motion, etc. However, there is still not a clear understanding of how all of these different local components are recombined to synthesize an apparently seamless visual scene. In my lab, we approach this problem by working at a level that is intermediate in complexity between single stimuli and complex scenes, i.e. at the level of processing multiple discrete stimuli. Our interest is in the ability of human observers to combine information from multiple stimuli across space and time. Specifically, we would like to know if the ability to process multiple stimuli can be predicted from what is known about the processing of single stimuli. To this end, we have extended the methods of classical psychophysics used in the study of single stimuli to the issue of integrating information from multiple stimuli.

The research in my lab can be classified into two broad categories: studies of visual search, and of visual motion processing. In visual search we are interested in building a model that characterizes the general problem of search; finding a particular piece of paper on a cluttered desktop or finding a person in a crowd. Of course, in the laboratory, we use simpler stimuli. A typical laboratory task involves finding a line of a particular orientation among shorter lines of random orientation. The real world equivalent of this task might be finding a needle in a haystack. Human performance in this task is well predicted by a simple model based on the known properties of orientation-selective detectors followed by a stage that computes the largest response among these detectors. What is most interesting is that the human visual system appears to combine the information from these multiple detectors in a manner that is close to the best performance that can be achieved in this task.

In the area of visual motion processing we have shown that current models of single motion detectors explain our ability to see brief motions, e.g. a dot moving along a short trajectory. However, when the motion trajectory is extended, as is typically the case in real life, human performance far exceeds that of individual motion detectors acting independently. We are currently investigating how local motion information may be combined to explain our enhanced ability to detect extended motion along a smooth path. Our results indicate that the visual system uses a predictive strategy: detectors early in the motion sequence might cue others later in the sequence that a motion stimulus is headed their way.

Collaborators: Suzanne McKee, James Coughlan, Laura Renninger, Doug Taylor, Tony Norcia, Alex Wade, Stefano Baldassi.