Models and Theory
Temporal precision in interception is surprisingly insensitive to image rate
Eli Brenner and Jeroen B. J. Smeets
People can swing a bat to hit a ball with a standard deviation in the timing of only about 7 ms. The temporal variability when hitting virtual targets is larger. We wondered whether this could be due to the intermittent presentation of virtual targets. To find out, we asked subjects to intercept moving targets by tapping on them with their index finger. The targets were 1.5 cm diameter disks that moved rightwards at about 40 cm/s across a large 120 Hz screen. Images of the smoothly moving target were presented for a single frame at 24, 30, 40, 60 or 120 Hz, with the luminance adjusted to the frame rate so that they all appeared to be equally bright (but some appeared to flicker somewhat). We defined timing error as the time between when the finger hit the screen and when the target centre passed the (horizontal) position at which it did so. Subjects hit the targets about 700 ms after they appeared (17 images at 24 Hz). The standard deviation in the timing error was about 12 ms, irrespective of the image rate. At low image rates this standard deviation is considerably smaller than the interval between the images.
Redundancy reduction: A possible explanation for the inhomogeneous patterns of antagonistic surrounds found in Middle Temporal (MT/V5) neurons
John A. Perrone and Richard J. Krauzlis
It is well established that motion sensitive MT neurons have an inhibitory region outside of their classical receptive field (crf) but no good explanation has been found for the fact that the inhibitory zones do not completely surround the crf but rather form inhomogeneous patterns around it [Xiao et al., 1997, Cerebral Cortex, 7(7), 662-677]. It is hard to reconcile these patterns with standard centre-surround spatial inhibition models. In the process of developing an image-velocity estimation model based on the outputs from a small number of MT neurons we discovered that these local patterns of inhibition could form part of an overall mechanism designed to reduce the amount of spatially redundant velocity signals fed to the next stage of motion processing (MST). We tested this idea by stimulating our velocity sensors and model MT pattern units [Perrone & Krauzlis, 2008, Journal of Vision, 8(9):1, 1-14] using small patches of moving dots similar to the Xiao et al., stimuli. With the inclusion of inhibition between adjacent MT units we were able to reduce the amount of redundant velocity signals and to replicate (and account for) the patterns of inhibitory zones found in the Xiao et al., sample of MT neurons.
Learning a spatio-temporal correlation
Devika Narain, Pascal Mamassian, Robert J. van Beers, Jeoren B. J. Smeets and Eli Brenner
We investigated whether participants could learn a space-time correlation without observing motion and how such learning depended upon experienced stimuli. Participants had to move a stylus to pass through a cued target position at the moment that the target was present. The target was only present for 100 ms, so participants had to anticipate when it would appear. Unbeknownst to the participants, when it appeared depended linearly on its horizontal position. Participants were divided into groups depending on how the stimuli were presented from trial to trial. For one group, stimulus positions were drawn independently from a Uniform distribution. For another group, stimuli changed position following a low-noise Random walk. In the latter case, good performance could be achieved by responding in accordance with the timing of the previous stimulus, without actually learning the spatio-temporal correlation. Participants quickly learned the correlation for the Uniform distribution. While performance was clearly better for the Random walk, when participants were later tested on uniformly distributed stimuli their performance showed that they had not actually learned the correlation. We conclude that participants can learn a space-time correlation without observing motion,
Impaired peripheral reaching and on-line corrections in patient DF: optic ataxia in visual form agnosia?
Stéphanie Rossit, Robert D. McIntosh, Stephen H. Butler, Larissa Szymanek, Stéphanie Morand, Ian G. Mackenzie, Hartmut Leuthold and Monika Harvey
An influential model of vision suggests there are two visual streams within the brain: a dorsal occipito-parietal stream which mediates action and ventral occipito-temporal stream which mediates perception. One of the cornerstones of this model comes from DF, a patient with visual form agnosia after bilateral ventral stream lesions. Despite her inability to recognize visual objects it has been shown that she can execute visually-guided actions towards them with a high-level of skill. These observations have been widely interpreted as demonstrating a double dissociation from optic ataxia, a condition observed after bilateral dorsal stream damage in which patients are unable to act towards objects that they can recognize. One major criticism against this view is that, while the visuomotor deficits in optic ataxia are typically manifest only with peripheral visual guidance, the integrity of action toward peripheral targets has not been closely examined in DF. Here we addressed this question, by asking DF to reach to stationary visual targets and also to perform fast on-line corrections in response to jumping targets, both in free and peripheral vision. Surprisingly, DF was remarkably inaccurate when reaching to peripheral stationary targets, but not when she was allowed to foveate the same targets. Moreover, she was also significantly impaired at correcting her reaches to target jumps, even in free vision. Our data suggest that DF presents visuomotor deficits surprisingly similar to those observed in optic ataxia, when tested under appropriately matched conditions.
Decision-making under time constraints supports sampling-based representation of uncertainty in vision
Marjena Popović, Máté Lengyel and József Fiser
Increasing body of psychophysical evidence supports the view of human perception as probabilistic inference that relies on representations of uncertainty about sensory stimuli and that is appropriate for statistically optimal decision making and learning. A recent proposal concerning the neural bases of these representations posits that instantaneous neural activity corresponds to samples from the probability distribution it represents. Since these samples are drawn sequentially, a crucial implication of such sampling-based representations is that precision of representing uncertainty will depend on the time available. To test this implication we created an orientation- matching task in which the subjects were presented several differently oriented line segments. We measured both subjects’ performance and their level of uncertainty as they matched the orientation of a randomly chosen element from the previously presented stimulus. We varied the stimulus presentation time trial-to-trial to influence the number of samples available before making a decision. We found that subjects’ performance and uncertainty judgment were correlated. Importantly, with decreasing the presentation time this correlation decreased significantly while the performance levels did not change. Thus, limiting the available time specifically influences the reliability of uncertainty representation, in agreement with sampling-based representations of uncertainty in the cortex.
Predicting Eye Movements in a Contour Detection Task
Udo Ernst, Nathalie van Humbeeck, Nadine Schmitt, Frouke Hermens and Johan Wagemans
Grouping local image elements into meaningful objects is a major task for the visual system. Hereby, an important process is contour integration, in which collinearly aligned local edges are merged into global contours. Recently, we developed a probabilistic model of contour integration which explains human contour detection behavior to a previously unprecedented degree [Ernst et al., PLoS Comp. Biol., in review]. Given this performance, we wondered whether the model might also explain the spatiotemporal dynamics of contour integration. As an indicator for these dynamics, we here focus on trajectories of eye movements during contour detection, hypothesizing that subsequent fixations preferentially visit ‘hotspots’ of neural activity which emerge during the integration process. In particular, we combine model simulations with recent experimental data [van Humbeeck et al., Perception, 40 EVCP Supplement, 192], in which eye movements were measured while human observers searched for a target contour embedded in a background of randomly oriented edges. The model was first used to predict potential locations for saccade targets, which were then compared to the real saccades. The target contour’s position was reliably predicted for stimuli in which also the majority of observers found its correct location, but not for stimuli where humans failed to identify the target. In addition, our model also predicted a large fraction of saccade targets visited before the actual contour was found, thus confirming both, our hypothesis, and the validity of our model.
Predicting Performance in Natural Scene Searches
Matthew Asher, Iain Gilchrist, Tom Troscianko and David Tolhurst
Completely natural scene search is a paradigm that cannot be directly compared to the typical types of search task studied, where objects are distinct and definable. Here we have look at the possibility of predicting the performance of humans for completely natural scene tasks, using a direct comparison of human performance against new and existing computer models of viewing natural images. For the human task, 25 participants were asked to perform a Target Absent/Target Present search task on 120 natural Scenes. The identical task was given to a selection of reproductions of existing computer processing techniques, including Feature congestion [Rosenholtz, Li, Mansfield, & Jin, 2005, SIGCHI ,pp. 761–770], Saliency [Itti and Koch, 2001, Journal of Electronic Imaging, 10(1), 161-169], Target Acquisition Model [Zelinsky, 2008, Psychological Review, 115(4), 787-835] and a new variation on the Visual Difference Predictor [To, Lovell, Troscianko, & Tolhurst, 2008, Proceedings of the Royal Society B-Biological Sciences, 275(1649), 2299-2308] These results lead us to conclude that in natural search tasks, the nature of both the Scene and the Target are important, and that the global influence of local feature groups can have an influence of the task difficulty.
Bayesian perception of orientation-dependent line length predicted by natural statistics
Adam Binch and Jim Stone
The perceived length of a line depends on its orientation, such that vertical lines are perceived as longer than horizontal lines of the same length (BJ Craven. (1993). Orientation dependence of human line-length judgements matches statistical structure in real-world scenes. Proc. R. Soc. Lond. B, 253:101-106), in a manner consistent with the natural statistics of line length (CQ Howe, and D Purves. (2002). Range image statistics can explain the anomalous perception of length. PNAS, 99:13184-13188). Here, we describe a method for estimating a Bayesian prior for the orientation-dependent perception of line length. The method was initially tested by recovering the known prior of a synthetic subject, and was then used to estimate the prior for each of five human subjects. These priors are in close agreement with the natural statistics of line orientation, suggesting that humans use the optimal prior for Bayesian inference of line length.