Crowding
Humans have a remarkable ability to recognize objects. We can recognize a chair as a chair despite variations in its position, size, or viewpoint, and we do it instantaneously. However, this process completely breaks down when objects are viewed in clutter (among other objects) in the periphery. The presence of closely spaced clutter turns a once-recognizable object into a jumbled mess. This crowding places a fundamental limit on perception. In typical viewing conditions, only objects near the center of the gaze survive crowding; we cannot recognize objects outside this “uncrowded window”. Try the demos to the left (from Pelli and Tillman, 2008). For each row, if you center your gaze on the red minus, you’ll find it impossible to recognize the middle object on the right. If you instead center your gaze on the gray plus/minus, the object will become recognizable.
We showed that the size of the uncrowded window limits the speed of reading (Pelli, Tillman, Freeman et al., 2007). During reading, we make a series of fixations, and at each fixation we harvest a small number of letters. Crowding limits the number of letters that we can recognize per glimpse, and thus determines reading speed. The uncrowded window of reading has important implications for the slow-reading that is characteristic of dyslexia, which may in part be due to excessive crowding.
Although crowding makes recognition impossible, some things survive crowding. In particular, we showed that change detection is immune to crowding. When asked to report changes between two crowded arrays of letters, observers could reliably report the change whether the letters were crowded or not. Thus, they could report changes between letters that were otherwise unrecognizable (Freeman and Pelli, 2007). Perhaps there is a low-level form of awareness that can detect changes between two object even without being able to say what either object is.
Currently, we are using probabilistic modeling to try to account for people’s mistakes during crowded object perception.