Thursday, March 21, 2019

Humans Trust Central Vision More Than Peripheral Vision Even in the Dark: humans have a default preference for information from the fovea even if this information is not veridical; filling-in precedes the estimation of confidence

Humans Trust Central Vision More Than Peripheral Vision Even in the Dark. Alejandro H.Gloriani, Alexander C.Schütz. Current Biology, March 21 2019.

•    Veridical information from the fovea is preferred under photopic viewing
•    Information missing in the scotopic foveal scotoma is filled in from the surround
•    Inferred information from the fovea is preferred under scotopic viewing
•    Content and properties of the foveal scotopic scotoma are hidden from awareness

Summary: Two types of photoreceptors in the human retina support vision across a wide range of luminances: cones are active under bright daylight illumination (photopic viewing) and rods under dim illumination at night (scotopic viewing). These photoreceptors are distributed inhomogeneously across the retina [1]: cone-receptor density peaks at the center of the visual field (i.e., the fovea) and declines toward the periphery, allowing for high-acuity vision at the fovea in daylight. Rod receptors are absent from the fovea, leading to a functional foveal scotoma in night vision. In order to make optimal perceptual decisions, the visual system requires knowledge about its own properties and the relative reliability of signals arriving from different parts of the visual field [2]. Since cone and rod signals converge on the same pathways [3], and their cortical processing is similar except for the foveal scotoma [4], it is unclear if humans can take into account the differences between scotopic and photopic vision when making perceptual decisions. Here, we show that the scotopic foveal scotoma is filled in with information from the immediate surround and that humans trust this inferred information more than veridical information from the periphery of the visual field. We observed a similar preference under daylight illumination, indicating that humans have a default preference for information from the fovea even if this information is not veridical, like in night vision. This suggests that filling-in precedes the estimation of confidence, thereby shielding awareness from the foveal scotoma with respect to its contents and its properties.


We investigated perceptual decision making under scotopic and photopic viewing. In two experiments, we found that a stimulus with a discontinuity in the scotopic foveal scotoma appeared as continuous, providing evidence for perceptual filling-in of the scotoma. We also found that observers preferred information from central vision, even when it was not veridical under scotopic viewing. This general preference for central vision indicates that humans are not aware of their scotopic foveal scotoma and that it is not taken into account for perceptual decision making.
Under daylight illumination, basic perceptual measures, such as acuity [13] or contrast sensitivity [14], peak at the fovea and decline in the periphery. In addition, the periphery is more vulnerable to crowding—i.e., spatial interference between neighboring elements [15]. Preferring information from central vision might be, therefore, a sensible strategy for decision making under ambiguity in photopic vision. This interpretation is supported by other foveal biases in photopic vision: stimuli with temporal and spatial uncertainty tend to be mislocalized toward the fovea [16], foveal brightness is extrapolated into the periphery [17], peripheral appearance is influenced by predicted foveal appearance [18, 19], and transsaccadic feature integration shows some overweighting of foveal information [20, 21]. However, the observed perceptual bias is not a useful strategy for scotopic vision, where the fovea does not contribute veridical information. Nevertheless, our finding is consistent with other perceptual phenomena where vision in the light and the dark is not calibrated well: perceived speed is underestimated in the dark [22], and the perception of white seems to require signals from cones [23]. Our results are at odds with a recent comparison of photopic and scotopic visual search [24], where eye movement statistics are affected by lighting condition in a qualitatively similar way as an ideal searcher [25], which has knowledge about the scotopic foveal scotoma. These divergent findings could point toward a general dissociation that the scotopic foveal scotoma is taken into account in eye movement control, but not in perceptual decision making. Alternatively, the divergent findings might be caused by different opportunities for learning in the two experimental paradigms. In the visual search task, observers experienced with every eye movement how visual input in the fovea and the periphery relate to each other and therefore had the opportunity to acquire the appropriate weighting of foveal and peripheral information. In the perceptual decision task of the current study, observers never experienced the same stimulus in the fovea and the periphery and therefore could not acquire the appropriate weighting during the experiment.
There are at least two ways how the perceptual bias could be caused in scotopic vision: first, the brain might use a simple heuristic that information from the fovea is more reliable than from the periphery and apply this heuristic to photopic and scotopic vision alike. However, a simple heuristic is unlikely, because humans can estimate uncertainty based on their actual perceptual performance instead of using simple cues, such as contrast or eccentricity in photopic vision [26]. Second, confidence might be assessed for each stimulus individually also in scotopic vision. In this case, our finding that biases in photopic and scotopic vision were similar, suggesting that confidence is assessed at a level of processing where information about the originating photoreceptor type is lost and perceptual filling-in is completed. Such a dissociation is quite likely, because rod and cone photoreceptors converge on the same pathways at the level of retinal ganglion cells [27, 28] and filling-in is preattentive [29] and takes place in visual cortex [9], while confidence in contrast seems to be represented only further downstream in parietal [30] and prefrontal cortex [31] and the striatum [32].
Several basic properties of visual processing, such as pupil size [33] or photoreceptor sensitivity [34], are directly adjusted to the light level during dark adaptation. Our results show that this is not the case for the relative weighting of foveal and peripheral information in perceptual decision making. However, other properties, such as rod-cone interactions [35] or spontaneous cortical activity [36], are controlled by a circadian rhythm rather than by light level. Since our measurements were taken during the day, it is possible that the relative weighting of foveal and peripheral information is also controlled by a circadian rhythm. In this case, the bias for foveal information should be reduced or even reversed at night but possibly in the same way for both scotopic and photopic viewing.
While there are only few and contradictory studies about filling-in of the scotopic foveal scotoma [6, 7, 10], more is known about filling-in at the blind-spot, where photoreceptors are absent because the axons of the ganglion cells exit the eye ball. Here, even complex visual patterns can be filled in from the surround [29], and humans are overconfident for this filled-in information [5]. Filling-in has also been observed for scotomata in the fovea caused by macular disease [37], and these patients need to acquire a new preferred retinal locus for fixation [38]. Our finding of a general preference for foveal information, irrespective of whether it is veridical or not, suggests that preferences in perceptual decision making might not necessarily shift to the preferred retinal locus in those patients, leading to suboptimal perceptual decisions.

No comments:

Post a Comment