Comments on Fred Kingdom's comments on my talk, by Alan Gilchrist.

Anchoring as applied to grating induction.

First I should say that Fred's description of the anchoring account of standard simultaneous lightness contrast (SLC) is excellent. As for applying the anchoring model to gradient induction, I recognize the difficulty. It is not immediately obvious how the display should be parsed into frameworks. But an appeal to the model as outlined in our 1999 review paper provides two rules relevant to the GI display:

RULE 1 Gradual luminance changes segregate frameworks.
RULE 2 T-junctions segregate the region above the T from the two regions flanking the stem.

I agree with Fred that GI differs from SLC in two basic ways: (1) the sinewave is repeated and (2) the target extends continuously across the light and dark inducing regions. The GI stimulus can be usefully simplified as shown in Figure 1d presented by Blakeslee and McCourt, and I will now apply the above three rules to this stimulus. Applying rule 1, we segregate the light and dark regions into separate frameworks divided by a gradient. As for T-junctions, there are two applications. First, there are T-junctions at each end of the horizontal target stripe. These group each end of the stripe to the inducing regions above and below it. But there are also T-junctions in the center of the display, with the gradient itself forming a fuzzy stem of the T. These T-junctions segregate the target stripe from the inducing regions. However, as Kardos (1934) has argued, a single homogeneous region cannot form a separate framework because, even though it may be segmented, it lacks articulation. In this case, Kardos applies a next-deeper-depth-plane rule; the target belongs to its background. The same analysis applies to Fred's stereo figure of the bar in front of the cylinder. I made the same point in my early experiments on depth and lightness (Gilchrist, 1980, page 533): "the lightness of an isolated target will be determined by its ratio to the noncoplanar context, even though this outcome would not occur with a more complex display."

I believe this provides an adequate basis for grouping the target stripe to the inducing regions above and below it. But we still must deal with the homogeneity of the target stripe. This homogeneity should make the two halves of the stripe appear equal in lightness. And I would argue that such an effect is indeed at work. One should expect a weaker induction effect in this case as compared to those in which the target stripe is broken, just what Blakeslee and McCourt show in their Figure 1, an outcome closely related to the Koffka-Benussi ring (Koffka, 1935) with, and without, a dividing line.

My demonstration pitting anchoring against intrinsic images.

First, Fred's memory of that demo is quite adequate.

Fred criticizes the presumed test because the luminance difference between the illuminated left half of the card and the non-illuminated background would produce a contrast effect, as in the Gelb effect, that would raise the lightness of regions on the left half of the card. I do not accept this criticism. Fred is mixing theories here. Let me clarify. I claim that this demonstration tests anchoring theory against my intrinsic image model (Gilchrist, et al, 1983). Contrast plays no role in that model. I believe that the heart of that model is classified edge integration, or the ability of the visual system to selectively integrate only the reflectance edges that fall between two targets. Thus a key prediction of that model is that the visual system can compute that the lightness of the target outside the spotlight (on the right) is five times higher than the target in the spotlight. It is never easy to establish that another person is wrongly criticizing their own model.

Furthermore, the outcome is rather indifferent to the luminance relationships that Fred cites. Increasing the average luminance of the far background either by increasing its reflectance or its illumination would have little or no effect on the result.

Let me add that, in addition to the challenge posed for the intrinsic image model, this demonstration also supports the anchoring model. Anchoring predicts that the target in the spotlight will be seen as at least slightly lighter (than the other target) because, while both targets have the highest luminance in their local frameworks, only the target in the spotlight has the highest global luminance. This is just the result we obtained (Gilchrist, et al, 1999) under the better controlled conditions of our laboratory test. Indeed many in the audience reported this outcome as well.

Do contrast theories need an anchor?

As for Fred's claim that in order to predict the relative lightness of targets in a display such as simultaneous lightness contrast, no anchor is needed, I agree.