Sensitivity to gaze-contingent spatial distortions in freely-viewed movies

Thomas Wallis, Michael Dorr and Peter Bex

Spatial resolution outside the fovea is progressively limited, and peripheral vision suffers from crowding, yet our subjective experience of the world is of uniform resolution. It has been suggested that this is due to learned representations of the statistical regularities in natural scenes. To probe these representations, we examined sensitivity to localised spatial distortions within which each pixel was shifted based on a bandpass noise process, preserving lower-order statistics but perturbing higher-order statistics. Four observers freely watched 8 hours of a blu-ray TV series on a low-latency gaze-contingent display. Distortions were presented at random intervals at one of four locations centred 4 degrees from the fovea and were spatiotemporally windowed in a Gaussian envelope (SD 1 degree and 80 ms). We varied the amplitude as well as the wavelength (0.25, 1, and 4 cpd) of the distortion. Observers each completed ~8,000 trials and were less sensitive to high- than to medium- and low-frequency distortions. We use reverse correlation to estimate the relationship between local image statistics at the target location and performance, both for retinotopic and spatiotopic statistics around the time of freely-made saccades.

Perceived numerosity in the peripheral visual field

Matteo Valsecchi, Matteo Toscani and Karl Gegenfurtner

Humans are able to estimate the number of elements in a cluster. Numerosity has been proposed as a basic visual feature given that it is susceptible to adaptation [Burr,Ross, 2008, Curr. Biol. 18,425-428] and numerosity detectors have been shown to emerge in a neural network coding sensory data [Stoianov, Zorzi, 2012, Nat. Neurosci. 194-196]. The appearance of basic visual features, in particular spatial frequency, is distorted in the periphery of the visual field [Thorpe et all., 1987, JOSA, 4,1606-1611]. Here we explored whether the perceived numerosity of peripheral elements is distorted. We presented eight observers with a reference central crowd of dots (N=30,60 or 90), forcing them to fixate its center using a gaze-contingent display. A test crowd of dots with variable numerosity was placed at 12 or 16 degrees of eccentricity. Observers indicated the more numerous crowd. PSE analysis showed that the numerosity of the clusters was underestimated by 7.5% at 12° and 17.6% at 16°. Our results are consistent with the idea that crowding limits the access of the numerosity detectors to some of the elements.

A Bayesian account of context-induced orientation illusions

Endel Poder

In classical tilt illusion, perceived orientation of a target object is repulsed from the orientation of surrounding context. Usually, it has been explained by an interaction between neural populations sensitive to different orientations. My earlier study (Poder 2009, Perception, 38 ECVP Supplement, 179) with crowded orientation stimuli showed a strong assimilation bias at about the same target-surround differences, opposite to the classical tilt illusion. In this study, I tried to clarify the mechanisms of these opposite biases. I presented the same stimuli (a central target and six surrounding Gabor patches) in central and peripheral locations of the visual field. For central presentation, stimuli with either high or low contrast were used. Orientation difference between the target and surround was varied. Absolute orientation was random. After the brief exposure, observers had to reproduce the perceived target orientation (by rotating a response patch). Both eccentricity and contrast manipulations indicate that reduced visibility of the target orientation increases the range of target-surround differences where assimilation bias is observed. It appears that perceived homogeneity vs. target pop-out determines the nature of illusion. A Bayesian model with competing (context-dependent) priors was built that produces qualitatively similar results. [Supported by Estonian Ministry of Education, project SF0180027s12]

How Recurrent Dynamics Explain Crowding

Aaron Clarke and Michael Herzog

In crowding, flankers impair target perception. For example, Vernier offset discrimination deteriorates when the Vernier is flanked by parallel lines. Pooling models explain crowding by averaging neural activity over both Vernier and flankers. Recently, however, it was shown that adding flankers can reduce crowding almost to baseline levels, contrary to predictions of pooling models. Here, we show that a Wilson-Cowan type model can explain both classical, local and recent, global aspects of crowding. The key feature of the model is spread of inhibitory neural activity across similar elements. For example, crowding strength decreases with more long flankers because these similar, long flankers inhibit each other dynamically and, thus, reduce inhibition on the dissimilar Vernier. Since the Vernier is similar to the equal-length flankers, it is inhibited. For this reason, and in accordance with psychophysical data, crowding does not vary with the number of equal-length flankers.

Does flanker identity affect perceived target position in crowding?

Funda Yildirim and Frans W. Cornelissen

A visual target is more difficult to recognize when other, similar, objects surround it. This is known as crowding. A recent model suggests that crowding is due to a combination of spatial and identity uncertainty [Van den Berg et al., 2012, J. Vision]. Here, we tested whether the identity of the flanking objects could also affect the perceived spatial position of a target, potentially adding to the spatial uncertainty. In our experiment, targets and flankers were presented in the peripheral visual field, and consisted of gabor patches. Two flankers that varied in orientation surrounded the target. An isolated patch with a fixed orientation was presented on the other side of fixation and served as a reference. Observers indicated whether the target or reference was closer to fixation. Pre-tests eliminated possible confounds. Orientation of an isolated target stimulus did not affect performance. As expected, neither did flankers presented outside the critical distance of crowding. In our main experiment, we found that flankers well within the critical spacing affected performance, and that flanker orientation modulated performance. We conclude that flanker identity affects the perceived position of the target.

Lesser crowding of letters generalizes across parafovea but not to vertical string orientation

Dušan Vejnovic and Suncica Zdravkovic

Contrary to the conventional bottom-up view [Pelli & Tillman, 2008, Nature Neuroscience, 11(10): 1129 - 1135], crowding can be dependent on object type. Namely, letters are shown to be less prone to crowding than some other symbols [Grainger et al, 2010, Journal of Experimental Psychology: Human Perception and Performance, 36 (3), 673-688]. This effect was attributed to perceptual learning. Using 2-AFC letter identification task, we examined whether the described effect generalizes to different parafoveal locations and to different string orientation. Results of Experiment 1 showed that letters are indeed less crowded than symbols at all tested locations in parafovea (left, right, above and below the fixation). Importantly, however, described letter advantage was only observed when the stimuli were shown in horizontal orientations. In vertical orientation there was no difference among letters and symbols. The effects of radial-tangential anisotropy were equal for both stimulus types. Results of Experiment 2 showed that radial-tangential anisotropy was canceled out when stimuli were presented at four diagonal positions in parafovea. This manipulation did not affect the selective advantage of the horizontally arranged letters in comparison to vertical letter strings and to symbols in either orientation. This research was supported by the Ministry of Education and Science of the Republic of Serbia (grant number: 179033)

Texture segmentation: Crowding can occur in detection tasks.

Ursula Schade and Cristina Meinecke

In visual crowding an irrelevant visual object can impair target recognition, even when it appears several degrees away from the target. Until now it is commonly agreed that crowding occurs only in target identification or target discrimination tasks, but not in target detection (e.g. Pelli et al., 2004, Journal of Vision (12), 4, 1136-1169). In a texture segmentation study, we investigated whether an irrelevant texture irregularity can impair target detection within a critical distance around the target. Data were also tested for anisotropic masking, which is considered as a 'litmus test'of crowding (Petrov et al., 2007, Journal of Vision (2), 7, 1-9). Furthermore it was investigated whether critical distances and anisotropic masking can be found not only when target and irrelevant irregularity are from the identical feature dimension, but also when they are from different feature dimensions. Results show that critical distances and anisotropic masking can be observed in a detection task. These two characteristics of crowding were found when the target and the irrelevant irregularity were from the identical feature dimension, as well as when they were from different feature dimensions.