Abstract
Objects in natural scenes are spatially bounded by sharp edges and comprise of narrow-band spatial-frequency components precisely aligned in phase. Early studies of feature integration using gratings suggested that integration across spatial frequencies is suboptimal. Here we re-examine this conclusion using a letter-identification task performed at fovea and 10° in the periphery. 26 lowercase letters with size twice the acuity at 10° were used for both fovea and periphery conditions. For each eccentricity, we determined the peak-tuning frequency (F) for the letters. We then measured the nominal letter contrast (Chung, Legge & Tjan, 2002) required to achieve 50% correct identification using octave-wide bandpass-filtered letters with center frequencies at 1/2F and 2F, as well as letters comprised of both the 1/2F and 2F components. The ratio of squared contrast sensitivity of the composite (1/C1/2F+2F)2 to the sum of squared contrast sensitivities of the components (1/C1/2F)2+(1/C2F)2 is 1.0 if the components are extracted independently but integrated optimally. A ratio lesser than 1.0 would indicate sub-optimal integration, while a ratio greater than 1.0 would indicate “super-optimal” integration, in that the visual system exploits correlations in noise and spatial uncertainty between spatial frequency components. We found that for both fovea and periphery conditions, contrast sensitivity for the composite letters were higher than that for the components. We also found that feature integration in the fovea is optimal, with an integration ratio of 1.05 ± 0.13. Surprisingly, this optimality is preserved in the periphery (integration ratio = 1.11 ± 0.19). This bears out the recent findings that for non-crowded letters, the first-order templates and second-order features recovered from the periphery are comparable to those recovered from the fovea (Nandy & Tjan, 2006), and that the spatial tuning for letter identification in the periphery is optimal (Chung, Legge & Tjan, 2002).
Supported by: NIH EY016391