Abstract
Objective:
Human visual system is able to recognize objects in large complexity variation. Despite such capability, little is known about the effects of complexity on object recognition. Here we studied the spatial frequency (SF) characteristics in identifying Chinese characters (CCs) of different complexity levels.
Method:
Stimuli were 150 frequently used CCs categorized into 3 complexity groups. Each character was digitally band-passed by 11 cosine log filters (bandwidth = 2 octaves, center frequency = 1.27 to 12.8 cycles/character in 0.1 log step). We measured contrast sensitivity for recognizing CCs of sizes 0.5°, 1°, and 2°. Peak SF (cycles/deg) and bandwidth (octaves) were plotted against character size in nominal character frequency (cycles/deg). A CSF ideal observer model (Chung et al., 2002 Vision Research
Results:
Log-log slopes of peak SF vs. size functions were 0.60±0.04 (M±SD), 0.67±0.02, and 0.72±0.05 for the low, medium, and high complexity groups. Bandwidth of the tuning functions was approximately 2 octaves for all complexity groups. Preliminary results from the CSF ideal observer analysis showed shallower slopes for the peak SF vs. size functions, but a similar trend for the bandwidth data compared with human performance.
Conclusions:
Peak SF of the tuning function did not scale perfectly with character size (log-log slopes < 1). The SF characteristics of CC recognition exhibited size-dependence, which differed across complexity groups. The ideal observer model utilizing human CSF and character-identity information failed to explain our data.