ALBERT

Description: We have shown that when searching for a disk in noise at one of 10 locations, the accuracy of the I st saccade is similar to that of the perceptual decision at the time of saccadic programming. The present study has two goals: 1) to test whether this linden - extends to a contrast-discrimination task without noise, and 2) to measure the effect of set size on the relation between saccadic and perceptual decisions. Methods: Three observers searched over a grey background (34.5 cd/sq m) for a bright disk (63.2 cd/sq m) among dim disks (54.1 cd/sq m) along the circumference of a circle (r = 5.9 deg.) centered on a fixation cress. Four set sizes (2, 4, 6, 12) were used. In the 1st condition, stimuli were presented for 1 sec. and observers used natural eye movements. We then measured the accuracy of the first saccade (% correct using a shortest-distance criterion). In the 2nd condition, observers fixated a central cross at all times and the stimulus duration was approx. 70 as less than the median latency of the first saccade in the 1st condition (saccadic programming time). We then recorded perceptual performance and discarded trials in which observers broke fixation. Results: For set sizes of 2, 4, 8, and 12, the mean d' across observers for the perceptual decision was 2.03, 1.96, 1.94, 1.71, respectively, while the mean d' of the first saccade was only 0.73, 1.40, 1.23, 1.17. Conclusions: Unlike detection of a disk in noise, for all observers and set-sizes, the perceptual accuracy at the time of saccadic programming is better than that of the lst saccade. For set-sizes of 4, 6, and 12, the amount of information available to the perceptual system relative to that available to the saccadic system is approximately constant (fixed do ratio). For these higher set sizes, the constancy in do across set size for both perception and saccadic decisions is consistent with a simple signal detection theory (SDT) model that processes noisy signals in parallel. However, for 2 observers, at a set-size of 2, saccadic targeting appears to be worse than the SDT model prediction, perhaps due to speed-accuracy trade-off.

Description: We have previously shown that humans can pursue the motion of objects whose trajectories can be recovered only by spatio-temporal integration of local motion signals. We now explore the integration rule used to derive the target-motion signal driving pursuit. We measured the pursuit response of 4 observers (2 naive) to the motion of a line-figure diamond viewed through two vertical bar apertures (0.2 cd/square m). The comers were always occluded so that only four line segments (93 cd/square m) were visible behind the occluding foreground (38 cd/square m). The diamond was flattened (40 & 140 degree vertex angles) such that vector averaging of the local normal motions and vertical integration (e.g. IOC) yield very I or different predictions, analogous to using a Type II plaid. The diamond moved along Lissajous-figure trajectories (Ax = Ay = 2 degrees; TFx = 0.8 Hz; TFy = 0.4 Hz). We presented only 1.25 cycles and used 6 different randomly interleaved initial relative phases to minimize the role of predictive strategies. Observers were instructed to track the diamond and reported that its motion was always coherent (unlike type II plaids). Saccade-free portions of the horizontal and vertical eye-position traces sampled at 240 Hz were fit by separate sinusoids. Pursuit gain with respect to the diamond averaged 0.7 across subjects and directions. The ratio of the mean vertical to horizontal amplitude of the pursuit response was 1.7 +/- 0.7 averaged across subjects (1SD). This is close to the prediction of 1.0 from vertical motion-integration rules, but far from 7.7 predicted by vector averaging and infinity predicted by segment- or terminator-tracking strategies. Because there is no retinal motion which directly corresponds to the diamond's motion, steady-state pursuit of our "virtual" diamond is not closed-loop in the traditional sense. Thus, accurate pursuit is unlikely to result simply from local retinal negative feedback. We conclude that the signal driving steady-state pursuit is not the vector average of local motion signals, but rather a more vertical estimate of object motion, derived in extrastriate cortical areas beyond V1, perhaps NIT or MST.

Description: Mulligan showed that the perceived direction of a moving grating can be biased by the shape of the Gaussian window in which it is viewed. We sought to determine if a 2-D pattern with an unambiguous velocity would also show such biases. Observers viewed a drifting plaid (sum of two orthogonal 2.5 c/d sinusoidal gratings of 12% contrast, each with a TF of 4 Hz.) whose contrast was modulated spatially by a stationary, asymmetric 2-D Gaussian window (i.e. unequal standard deviations in the principal directions). The direction of plaid motion with respect to the orientation of the window's major axis (Delta Theta) was varied while all other motion parameters were held fixed. Observers reported the perceived plaid direction of motion by adjusting the orientation of a pointer. All five observers showed systematic biases in perceived plaid direction that depended on Delta Theta and the aspect ratio of the Gaussian window (lambda). For circular Gaussian windows Lambda = 1), plaid direction was veridically perceived. However, biases of up to 10 deg. were found for lambda = 2 and Delta Theta = 30 deg. These data present a challenge to models of motion perception which do not explicitly consider the integration of information across the visual field.

Description: Viewing a scene through an optical window provides observers with numerous visual properties. In order to create a 'virtual window' that is perceptually compelling, it must be determined which properties are most critical to preserve. We have examined several properties, both static and dynamic, and will discuss which have the greatest impact on apparent realism (and user performance).

Description: We have been developing a simplified spatial-temporal discrimination model similar to our simplified spatial model in that masking is assumed to be a function of the local visible contrast energy. The overall spatial-temporal sensitivity of the model is calibrated to predict the detectability of targets on a uniform background. To calibrate the spatial-temporal integration functions that define local visible contrast energy, spatial-temporal masking data are required. Observer thresholds were measured (2IFC) for the detection of a 12 msec target stimulus in the presence of a 700 msec mask. Targets were 1, 3 or 9 c/deg sine wave gratings. Masks were either one of these gratings or two of them combined. The target was presented in 17 temporal positions with respect to the mask, including positions before, during and after the mask. Peak masking was found near mask onset and offset for 1 and 3 c/deg targets, while masking effects were more nearly uniform during the mask for the 9 c/deg target. As in the purely spatial case, the simplified model can not predict all the details of masking as a function of masking component spatial frequencies, but overall the prediction errors are small.

Description: Although numerous studies have examined the relationship between smooth-pursuit eye movements and motion perception, it remains unresolved whether a common motion-processing system subserves both perception and pursuit. To address this question, we simultaneously recorded perceptual direction judgments and the concomitant smooth eye movement response to a plaid stimulus that we have previously shown generates systematic perceptual errors. We measured the perceptual direction biases psychophysically and the smooth eye-movement direction biases using two methods (standard averaging and oculometric analysis). We found that the perceptual and oculomotor biases were nearly identical suggesting that pursuit and perception share a critical motion processing stage, perhaps in area MT or MST of extrastriate visual cortex.

Description: Current models of smooth pursuit eye movements assume that it is largely driven by retinal image motion. We tested this hypothesis by measuring pursuit of elliptical motion (3.2s, 0.9 Hz, 1.4 deg x 1.6 deg, 4 randomly interleaved phases) of either a small spot ("real" motion) or of a line-figure diamond viewed through apertures such that only the motion of four isolated oblique line segments was visible ("virtual" motion). Each segment moved sinusoidally along a linear trajectory yet subjects perceived a diamond moving along an elliptical path behind the aperture. We found, as expected, that real motion produced accurate tracking (N = 2) with mean gain (over horizontal and vertical) of 0.9, mean phase of -6 deg (lag), mean relative phase (H vs V) of 90 +/- 8 deg (RMS error). Virtual motion behind an X-shaped aperture (N= 4 with one naive) yielded a mean gain of 0.7, mean phase of -11 deg, mean relative phase of 87 +/- 15 deg. We also measured pursuit with the X-shaped aperture using a higher segment luminance which prevents the segments from being grouped into a coherently moving diamond while keeping the motion otherwise identical. In this incoherent case, the same four subjects no longer showed consistent elliptical tracking (RMS error in relative phase rose to 60 deg) suggesting that perceptual coherence is critical. Furthermore, to rule out tracking of the centroid, we also used vertical apertures so that all segment motion was vertical (N = 3). This stimulus still produced elliptical tracking (mean relative phase of 84 +/- 19 deg), albeit with a lower gain (0.6). These data show that humans can track moving objects reasonably accurately even when the trajectory can only be derived by spatial integration of motion signals. Models that merely seek to minimize retinal or local stimulus motion cannot explain these results.

Description: We have previously shown that, during simulated curvilinear motion, humans can make reasonably accurate and precise heading judgments from optic flow without either oculomotor or static-depth cues about rotation. We now systematically investigate the effect of varying the parameters of self-motion. We visually simulated 400 ms of self-motion along curved paths (constant rotation and translation rates, fixed retinocentric heading) towards two planes of random dots at 10.3 m and 22.3 m at mid-trial. Retinocentric heading judgments of 4 observers (2 naive) were measured for 12 different combinations of translation (T between 4 and 16 m/s) and rotation (R either 8 or 16 deg/s). In the range tested, heading bias and uncertainty decrease quasilinearly with T/R, but the bias also appears to depend on R. If depth is held constant, the ratio T/R can account for much of the variation in the accuracy and precision of human visual heading estimation, although further experiments are needed to resolve whether absolute rotation rate, total flow rate, or some other factor can account for the observed -2 deg shift between the bias curves.

Description: There has long been qualitative evidence that humans can pursue an object defined only by the motion of its parts. We explored this quantitatively using an occluded diamond stimulus. Four subjects (one naive) tracked a line-figure diamond moving along an elliptical path (0.9 Hz) either clockwise (CW) or counterclockwise (CCW) behind either an X-shaped aperture (CROSS) or two vertical rectangular apertures (BARS), which obscured the corners. Although the stimulus consisted of only four line segments (108 cd/square m) moving within a visible aperture (0.2 cd/square m) behind a foreground (38 cd/square m), it is largely perceived as a coherently moving diamond. The inter-saccadic portions of eye-position traces were fit with sinusoids. All subjects tracked object motion with considerable temporal accuracy. The mean phase lag was 5 deg/6 deg (CROSS/BARS) and the mean relative phase between the horizontal and vertical components was +95 deg/+92 deg (CW) and -85 deg/-75 deg (CCW), which is close to perfect. Furthermore, a chi-square analysis showed that 56% of BARS trials were consistent with tracking the correct elliptical shape (p is less than 0.05), although segment motion was purely vertical. These data disprove the main tenet of most models of pursuit: that it is a system that seeks to minimize retinal image motion through negative feedback. Rather, the main drive must be a visual signal which has already integrated spatiotemporal retinal information into an object-motion signal.

Description: A template model of human visual self-motion perception, which uses neurophysiologically realistic "heading detectors", is consistent with numerous human psychophysical results including the failure of humans to estimate their heading (direction of forward translation) accurately under certain visual conditions. We tested the model detectors with stimuli used by others in single-unit studies. The detectors showed emergent properties similar to those of MST neurons: (1) Sensitivity to non-preferred flow; Each detector is tuned to a specific combination of flow components and its response is systematically reduced by the addition of nonpreferred flow, and (2) Position invariance; The detectors maintain their apparent preference for particular flow components over large regions of their receptive fields. It has been argued that this latter property is incompatible with MST playing a role in heading perception. The model however demonstrates how neurons with the above response properties could still support accurate heading estimation within extrastriate cortical maps.