UC Berkeley

Object recognition is a fundamental function of the human visual system. Our visual system must recognize objects under a wide variety of environmental conditions and despite inherent noise in the nervous system. How the brain accomplishes such accurate and reliable performance on object recognition tasks is widely studied but poorly understood. Less understood still is how the brain accomplishes object recognition in a broader spatial and temporal context, as many studies focus on studying the recognition of objects in isolation. Objects in the real world are more often seen and recognized in both a surrounding visual environment, and for more than one moment in a dynamic world. Here we address these questions by examining the phenomena of crowding and serial dependence. Crowding is an interaction of spatial context; it is the inability to recognize an object when it is surrounded by other similar objects, which would otherwise be resolved in isolation. Serial dependence is an interaction of temporal context; a target appears more similar to a previously seen stimulus. First, we ask what types of information survive crowding and what types of representations can crowd one another. Through computational modeling, we present evidence that crowding is a perceptual decision level phenomenon that occurs when we must make examine a distributed representation with limited attentional resources. Next, in a behavioral experiment, we show that lighting information in the context of a scene can influence how nearby object representations interact in crowding. Finally, we show that temporal context can influence the perception of even high level dynamic stimuli like point-light walkers through the mechanism of serial dependence. Together, these experiments show that both spatial and temporal context play an important role in object recognition and that the brain uses high level object representations that can interact in both space and time to accomplish recognition.