Abstract
Previous noise masking studies showed that only gratings or band-pass filtered letters exhibit perfect channel scaling (log-log slope = 1.0 for channel frequency vs. stimulus peak frequency). Recognition of unfiltered letters requires frequency channels higher than the stimulus frequency, following a 2/3 slope. This deviation is explained by the increasing use of high frequency edges for recognition. Here we further investigated channel scaling in letter recognition using Sloan letters (stroke frequency (SF) = 2.0 strokes/char) and simple and complex Chinese characters (CC1/CC6, SF = 2.2/5.5 strokes/char, respectively). Band-pass noise masking was used to determine character recognition channels at different stimulus sizes. The channel frequency for Sloan/CC1 scaled along the 2/3 slope line, but for CC6 followed the 1.0 slope line despite of the sharp stimulus edges, indicating that Sloan and CC1 required a higher channel than did CC6 at the same stroke frequency. However, the more complex CC6 was much larger in size than Sloan/CC1 at the same stroke frequency. If the same channel with a fixed receptive field size inspects all these stimuli, and if it can collect enough samples over the larger CC6, the channel may be too coarse to collect enough samples over the much smaller Sloan/CC1. The deviation of Sloan/CC1 channels from perfect scaling thus may reflect the need to use smaller receptive fields (thus higher frequency channels) to obtain more spatial samples. To test this hypothesis, we measured the channel for CC6 recognition with randomly placed holes. The channel started to deviate from the 1.0 slope line with 35% hole coverage and approached the 2/3 line with 70% hole coverage while the stimuli were still recognizable. These results suggest that because channel frequency is related to the receptive field size in space, sufficient number of spatial samples collectable from an object may determine channel scaling.
Natural Science Foundation of China Grant 30725018.