Abstract

A 2-D model consisting of an array of self-similar Gabor filters (bandwidth = 1 octave) followed by a power function contrast transducer (exponent = 0.5) and response pooling across spatial frequency and orientation using the Quick summation rule (exponent Q = 2.5) was shown to simulate accurately human estimates of apparent contrast [ARVO (1988)]. However, magnitude estimation experiments indicate that the contrast transducer exponent could be anywhere between 0.3 and 0.7. To test the robustness of the original estimates of filter bandwidth and Q, model estimates were compared to contrast matching data when contrast transducer function exponents ranged from 0.1 to 0.7. Results demonstrated that filter bandwidth estimates remained constant, but the Q value required to give the best fit between model and data was found to be a function of the contrast transducer exponent. Further tests of the model when the Gabor filters were replaced by different of Gaussians or a difference of three Gaussians showed that model performance was still better for bandwidths near 1 octave and showed exactly the same functional relationship between Q and the contrast transducer exponent. Thus, while model performance is relatively independent of the exact mathematical form of the filters, the final specification of Q for response pooling requires a more accurate description of the contrast transducer function.

© 1988 Optical Society of America