O letter channels, where art thou?

At VSS 2001, we reported that the spatial-frequency tuning characteristics for letter identification could be accounted for by the product of a function describing letter-identity information vs. spatial frequency and an observer's contrast sensitivity function (CSF). This result obviates the need to invoke specific channels for letter identification (e.g. Solomon & Pelli, 1994). Here, we tested our theory by adding a pedestal white noise to the noise-masking paradigm used by Solomon & Pelli. A sufficient amount of pedestal noise can render an observer's intrinsic noise ineffective and hence bypasses the CSF. Our theory predicts that tuning functions should appear broader when the CSF is bypassed. In contrast, the letter-channel model predicts no difference. To test our hypothesis, we measured contrast-energy thresholds for identifying 1.6° letters embedded in luminance noise. The masking noise was produced by high- or low-pass filtering a white noise (rms contrast of 20%) at one of nine cutoff frequencies (including a no-noise and a unfiltered white noise condition). The spatial tuning function of a letter "channel" is the derivative of the threshold vs. cutoff-frequency functions so obtained. Then, we re-measured contrast-energy thresholds with an additional white noise (pedestal) added to the stimulus and the masking noise. This pedestal noise had an rms contrast of 9%, several times higher than the intrinsic noise of our observers for the testing condition. Consistent with our prediction, observers showed an increase in tuning bandwidth in the presence of the pedestal noise (from an average full-width-at-half-height of 1.4 to 2.7 octaves). Averaged peak tuning frequency also decreased from 3.5 to 2.7 c/letter. These findings are consistent with the idea that the observed letter "channels" are the results of an interaction between letter-identity information across spatial frequencies and the visual system's limit in spatial resolution.