Complex channels become more complex: Modeling a contrast adaptation process

Abstract

We have found a new (at least to us) kind of contrast adaptation in which differences between contrasts near the adapted level are harder to perceive than differences between contrasts further away. In one experiment, the observer is briefly adapted to a grid of Gabor-patch elements at some contrast level (C₀). We then probe the adapted state using a striped element-arrangement pattern where the Gabors in alternating rows (or alternating columns) change contrast level to C₁ while the Gabors in the remaining rows (or columns) change to C₂. If C₁ is an increment in contrast relative to C₀ (that is, if C₁=C₀+ΔC) and C₂ is a decrement in contrast relative to C₀ (that is, if C₂=C₀-ΔC), the orientation of the striped pattern is very hard to identify. However, the orientation is easy to identify if the same size difference (2*ΔC) does NOT cross C₀ (for example, if C₁=C₀+ΔC₁ and C₂= C₀+3*ΔC).

We can model this contrast-controlled adaptation process by inserting an additional linear filter (F₂) and rectification-type nonlinearity (N₂) into a complex channel (F₁N₁F₃), resulting in a F₁N₁F₂N₂F₃ structure (which we call a Buffy channel). F₂ is a small blur circle, so the output of it is a measure of local contrast. N₂ is adaptable: the zero-point of N₂ adapts to be equal to the recent time-averaged contrast (at that position), so the output is (approximately) zero if the current contrast equals the recent contrast (at that position).

Predictions from Buffy channels agree with experimental results.

Footnotes

 Supported by NIH grant EY08459