Computing stereo channels from masking data.

The detection of stereoscopic depth in random-dot patterns that have been spatially band-pass filtered is adversely affected by the addition of noise at spatial frequencies in the neighbourhood of the frequencies present in the stereogram. This elevation of threshold is generally termed masking and recent data have been interpreted as evidence for a pair of spatial-frequency-tuned stereo "channels" whose peak spatial frequencies are either at 3 and 5 c/deg or 2.5 and 7 c/deg. This interpretation was re-examined. In particular, we have studied how the characteristics of masking interactions might be affected by taking account of the presence of an initial modulation transfer function (including the optical m.t.f. of the eye) that precedes the stage at which signal and mask interact. Using this approach, we reach the conclusion that the peak of the internal masking function for stereo detection coincides with the signal spatial frequency over the whole range tested (1.7-11.6 c/deg). We conclude that the recent data of Yang and Blake [(1991). Vision Research, 31, 1177-1189] are consistent with a multiple channel model in much the form proposed by Julesz and Miller [(1975). Perception, 4, 125-143]. The analysis presented in this paper has general implications for the interpretation of masking studies in spatial contrast vision.