Abstract
At ARVO 1997, Majaj, Kurshan, & Pelli used critical-band masking to show that letter identification is not scale invariant. They found that channel frequency grows as the 2/3 power of line-frequency, a unit for measuring target spatial frequency. (Scale invariance would produce a value of 1.) At the same meeting, Chung & Legge presented data on contrast sensitivity for identifying filtered letters of different sizes and reported the same result. It is remarkable that the two groups, one using filtered noise, the other filtered letters, arrived at the same quantitative finding. What is the cause of this lack of scale invariance for letter identification? We measured the foveal contrast sensitivity function (CSF) of a participant in the study of Chung & Legge with sine-wave gratings (as opposed to letters). An ideal observer with this CSF as the front-end followed by additive white noise was used to perform the same letter identification task with filtered letters as in Chung & Legge. We found that the “channel” frequency for the ideal observer grows as the 0.69 power of the target spatial frequency, in excellent agreement with human data. We conclude that the scale dependence of letter identification is due to an interaction between the distribution of letter-identity information across spatial frequencies and the visual system's limit in spatial resolution — not a result of any strategic feature selection by the human observer.
Supports: NEC Research Institute (Tjan), NIH Grant EY02934 (Legge), and EY12810 (Chung).