Purchase this article with an account.
Gerald Sun, Susana T. L. Chung, Bosco S. Tjan; Mechanisms of crowding and learning to “uncrowd”. Journal of Vision 2008;8(6):438. doi: https://doi.org/10.1167/8.6.438.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Nearby patterns adversely affect identification of an eccentrically presented target in a phenomenon known as crowding. In peripheral vision, perceptual learning improves single-target identification (Gold, Bennett, Sekular, 1999; Chung, Levi, Tjan, 2005) and reduces the extent of crowding (Chung, 2007). However, the mechanisms of crowding and of the “uncrowding” effect following learning remain unknown. We used a noise-masking paradigm, combined with perceptual learning, to uncover these mechanisms. Pre- and post-tests measured threshold contrast energy (at 50% correct identification) versus noise energy (EvN function) for identifying isolated and flanked letters at 10° in the lower visual field; letter size was 2.5x a subject's acuity. Flanking letters, when present, were at 33% Weber contrast and placed on both sides of the target letter at a center-to-center distance of 1 x-height. Four levels of static white luminance noise (rms contrast = 0, 7.9%, 12.6%, 20%) were added within the largest bounding box of the target letter. Six days of training immediately followed the pre-test at the same letter size and eccentricity, but with a target-flanker separation of 0.8 x-height. Training was effective (+10% correct, N=3). Pre-test EvN functions indicated that crowding both reduced sampling efficiency to 25% of the value of the isolated-letter condition and increased an observer's intrinsic noise by the equivalent of adding a white noise of 19% rms contrast to the target. Following training, the ratio of post- to pre-test efficiency was practically unchanged (1.1), but the flanker-induced intrinsic noise was reduced to 30% of its pre-training value. Our findings indicate that (1) both the precision and computational strategies of the visual system are severely affected by crowding, and (2) perceptual learning reduces intrinsic noise without affecting sampling efficiency. Thus, perceptual learning moderates crowding by partially restoring some of the lost precision but makes no improvement on computational strategies.
This PDF is available to Subscribers Only