[back]

Suppressive Effect of Sustained Low-Contrast Adaptation followed by Transient High-Contrast on Peripheral Target Detection
Farshad Moradi, Shinsuke Shimojo, Christof Koch

Filling-in can be induced by high-contrast edge adaptation, or after prolonged adaptation to a peripheral low-contrast object (Troxler fading). Adaptation to sustained low-contrast vs. adaptation to transient high-contrast suggests synergy between contrast and edge adaptation, but the possible interactions are not well understood. We observed that briefly increasing the contrast of a peripheral low-contrast object after a few seconds of strict fixation elicits disappearance of the object, resulting in perceptual filling-in of the location with the surround (Figure 1a). After a short time usually around one second the object reappears. Hence, following sustained adaptation to a low-contrast target, transient high-contrast stimulation can induce perceptual disappearance.

The induced disappearance illusion was equally strong when the contrast of the high-contrast flash was inverted (Figure 1b). Therefore, the disappearance of the target can not be explained by retinal light adaptation. We used Gabor patches to study induced disappearance in terms of edge adaptation and found that presenting a low-contrast Gabor patch (2cpd, 5deg eccentricity, contrast = 4%) for 8 seconds and then flashing a 20-30ms high-contrast patch over it could elicit the perceptual disappearance of a subsequent low-contrast stimulus, whereas neither low-contrast adaptation nor high-contrast flash alone had any considerable effect (p<.00001).

In other experiments we found: a) adaptation and induction are phase-insensitive, b) the effect transfers between eyes, c) suppression is selective for orientation, and d) the induction by the transient high-contrast Gabor patch could be transferred to another previously adapted location up to a few degrees. Results indicate synergy between contrast and adaptation through a non-linear interaction between rapid gain adjustment to transient change and adaptation to sustained spatial patterns. Findings are compatible with non-local mechanisms at the intermediate cortical levels of visual processing. Considering the similar characteristics in a wide variety of experimental manipulations, the same mechanisms may also underlie suppression of object boundaries in illusions such as motion-induced blindness or fading induced by visual transient.

We proposed a model in terms of contrast gain and offset based on the idea of optimal neural encoder. Assume that the mapping (alternatively the psychometric curve) is monotonic in form of f(input/gain - offset), where f is the standard normal cumulative distribution function, and gain reflects the standard deviation of the input, and the observer has some inherent internal noise which is independent of the input (Figure 2a). Discriminability of two levels of stimulus intensity is inversely proportional to the slope of f at those values. For efficient coding (in statistical sense), discriminability should be high for events that occur with high frequency, and low for low-frequency events. That is, the mapping (offset, gain) should conform to the distribution (mean, variance) of the inputs. As the distribution is not a priority fixed, the ideal observer should estimate and dynamically update distribution parameters. For inputs around the mean, probability of the input given the distribution is around its maximum. Therefore, the likelihood of distributions is minimally affected. Consequently, estimation of the optimal offset for neural code requires temporal integration. In contrast, a sudden increase in the range of stimulus intensities dramatically alters posterior probabilities and rapidly modulates the gain.

In this framework, we propose that sustained low-contrast adaptation gradually increases the offset and the gain, elevating the detection threshold. Then, induction reduces the gain without effectively affecting the offset, resulting target contrast to fall below the range of intensities encoded effectively by neurons, as schematically illustrated in Figure 2b. The model is consistent with electrophysiological data and psychophysical experiments that showed that detection threshold (which reflects offset) but not discrimination threshold (which reflects gain) increase after prolonged adaptation to low-contrast stimuli.

 

Figure 1. a, b) Induction of disappearance by brief presentation of a high-contrast stimulus with the same (a) or opposite (b) polarity after adaptation to a low-contrast pattern. c) Reversing the order of the sequence removes the effect.

Figure 2. The hypothetical stimulus-response curve for neurons and its modulation after adaptation. a) the stimulus-response curve before adaptation. This curve is optimal for intensities around m1 (indicated by a small arrow above the graph). However, this curve is not optimal for stimulus intensities around m2, and the response for x is saturated. b) adaptation may improve coding efficacy by modulating the offset (threshold) of the stimulus-response curve in the case that the mean input intensity is m2 (given a constant variance), or by reducing the gain (slope) when the mean is m1 but some samples are as high as x (increased variance). These conditions respectively correspond to sustained adaptation to low-contrast vs. induction (adaptation to high-contrast). Notably, modulation of both offset and gain has a combinatory effect, resulting m2 to fall below the threshold. c) Biological plausibility: the input current vs. firing rate curve for a leaky integrate-and-fire model neuron with refractory period. Shunting-inhibition elevates the offset, where as increasing the spike-threshold (via a hyperpolarizing current) modulates the gain. Again, there is a large combinatory effect when both gain and offset change.