Abstract
Background. Graham & Nachmias (1972) found minimal subthreshold summation of a 1st plus 3rd harmonic, strongly supporting a multiple channel model against an ideal observer template model. However there is mounting evidence for a template model in the suprathreshold regime. We repeated their experiment in the presence of noise with the goals of determining: (i) the observer's template, (ii) whether an ideal observer or a multiple channel model was operating for these suprathreshold stimuli, and (iii) the ratio of internal systematic vs. random noise.
Methods. The noise was the sum of 1st through 7th harmonics of a 1 c/deg grating. Six cosine test patterns were used consisting of the 2nd, 6th, and four combinations of the 2nd and 6th harmonics. We used the method of constant stimuli with 4 test contrasts and 4 responses. A classification template was obtained by linear regression of the responses on the cosine noise amplitudes. Each run was repeated (double pass) to measure the ratio of systematic vs. random noise. A second method for measuring this ratio is based on the d' ratio of the human observer to a template observer (predicted ideal response to the stimulus using the classification template).
Results & Discussion. We found that some templates can be formed efficiently, improving results beyond that of the multiple channel prediction by Graham & Nachmias. This provides evidence for a template observer in suprathreshold detection tasks. However, the fact that this was not found in all cases suggests that observers had difficulty creating an efficient classification template for some of the compound stimuli. The double pass analysis implied that internal systematic noise (mismatched template rather than random noise) was the main factor controlling thresholds.
Graham, N. & Nachmias, J. (1972). Detection of grating patterns containing two spatial frequencies: A comparison of single channel and multiple channel models. Vision Research, 11, 251–259.