Abstract
Purpose: To study the degree of abstraction from an object's stimulus image to its internal representation, and the role of familiarity in this abstraction. To effectively study this abstraction, objects are either imbedded in luminous noise or partially occluded. Method: In a same-different matching task, two images were sequentially presented (1st image: 1s, mask: 0.2s, 2nd image: 0.25s, mask). Subjects decided if the two images were the same or different letters. Each letter was displayed in one of five noise conditions. The noise could either be luminous noise or fence occluders, presented in a blocked design. The luminous noise was additive white Gaussian pixel noise and the fence occluders were parallel bars. In each block, therefore, there were 25 conditions. The subjects were all native English speakers, to whom the letters were either familiar (English letters) or unfamiliar (Chinese characters). Results: (1) In the high noise range, those conditions where both images shared the same noise level gave rise to high accuracy, suggesting that subjects used image-based matching. (2) In the low noise range, the lower the noise level, the higher the accuracy, suggesting that subjects could perceptually complete the letters regardless of familiarity. (3) In the medium range of noise, similar results as in (2) were obtained for the (familiar) English letters, suggesting that subjects could completely recover the letters in the internal representation. (4) Most interestingly, in the medium noise range and with (unfamiliar) Chinese characters, those conditions with the noise level of the 2nd image only 1 or 2 steps lower, but not further, than the 1st image gave rise to high accuracy, suggesting that subjects achieved a limited degree of abstraction by perceptual completion.
Support: NSF IBN-9817979.