|
Promoting vision research and its applications |
|
||||||||
|
AVA 5th Christmas Meeting Wednesday 20th December 2000 University of Surrey
Talks Orientation Mechanisms: A model Models of frequency- and orientation-selective filtering in mammalian visual cortex adequately describe the first-stage of analysis in visual cortex. Models of processing beyond the "simple" cells, the second stage mechanisms, are less clear. For example, second stage energy mechanisms have been proposed that give the correct orientation for a sinusoidal grating image, but do this as a uniform field, failing completely to account for the percept of light and dark bars. The model proposed here estimates orientation accurately and places the "edges" and centres of light and dark bars of a grating in good agreement with the positions perceived by human subjects. The model explains the checkerboard appearance of two sinusoids at ± 45 degrees (additively superimposed to form a "plaid"), it provides mechanisms for orientation pop-out, and it reproduces the Mach band effect. The proposed model uses both the energy and the phase responses of the first stage filters (here we consider static, monochromatic, monocular processing at a single scale). The model goes beyond simple "edge" and "bar" features to detect and quantify the orientation of local patterns with higher-order rotational symmetries. The processing implied by the proposed model is biologically plausible, simple, robust, and unifies the processing of energy and phase mechanisms. The second-stage processing results in a family of feature maps that have implications for the understanding of later processing eg pop-out and texture segmentation, and the nature of the processing suggests "complex" cells with a variety of properties. The talk will be illustrated by numerous examples of the analysis of test and real-world images, with indications of the expected receptive fields of some of the second stage mechanisms.
What direction corresponds to `depth' for stereoscopic vision? Binocular disparities provide information about depth differences, but the axis along which the visual system measures depth has not been determined. Here, we present evidence suggesting that the stereoscopic visual system varies the direction defined as the `depth axis' depending on the surface being viewed. Subjects judged whether the central column of a 7 by 7 grid of bright dots (6 degrees square, 2 arcmin dots) was in front of or behind the plane of the grid. The grid was slanted about a vertical axis (disparity gradient +/- 0.1, constant within one run). Three conditions were randomly interleaved: the central column of dots was displaced (i) only in the right eye, (ii) only in the left eye or (iii) by an equal and opposite amount in each eye. Stereoacuity thresholds were determined from at least 1400 trials per condition. Thresholds varied systematically with the slant of the grid. In condition (i), thresholds were lower when the right hand edge of the grid had a crossed disparity than for the opposite slant. The reverse was true for condition (ii). Thresholds in condition (iii) were intermediate between (i) and (ii). The results are consistent with the possibility that the stereoscopic visual system is primarily sensitive to displacements perpendicular to the plane of a local surface, not to displacement along an axis relative to the head. A surface-based representation of depth such as this would be relatively unaffected by translations or rotations of the head.
Motion Segmentation and Transparency: A Computational Analysis and Some When multiple motion directions are present within the same region of the visual field, the phenomenon of motion transparency can be observed. This can be regarded as an intermediate state between the breakdown of scene segmentation and the complete absence of meaningful motion information (noise) - and therefore is of high theoretical and biological significance.
It takes time to bind: visual features are poorly localised in brief Exposures Signal Detection Theory (SDT) asserts that sensory analysis is limited onlyby noise, and not by the number of stimuli analysed. To test this claim, we measured the accuracy of visual search for a single tilted element (the target) amongst seven horizontal elements (distractors) using several different exposure durations, each terminated by a random noise mask. In the uncued condition each element was a potential target. In the cued condition only two were. SDT predicts that location errors should be evenly distributed amongst all distractors. For long (e.g. 5.0 s) exposures, this prediction was confirmed and SDT could simultaneously fit uncued and cued accuracies. For short (e.g. 0.1 s) exposures, errors were concentrated amongst distractors adjacent to the target and, unless modified to account for this, SDT underestimated the difference between uncued and cued accuracies. Therefore when the time available for search is brief, features such as orientation can be seen, but their positions can only be roughly estimated. That is, it takes time to bind features to their positions. The ghosts of a Christmas target, present, future and past How does immediate context effect search for a specific target? We used a simple visuomotor task to explore this question. Observers sequentially clicked through a series of targets (e.g., letters A-J) while we manipulated the availability of information ahead or behind of the current target. To manipulate retrospective information, targets could remain once clicked or they could vanish. To manipulate prospective information, the whole sequence could be shown or we could shuffle or mask targets ahead in the sequence. Our findings indicate that both "where you've been" and "where you're going" have a large impact on the efficiency of finding a particular target in this task. Interestingly, retrospective and prospective aspects of search appear to interact, so that search performance is poorest when old targets are still present and new targets are not available for response planning.
Selective use of visual information for action We generally have the impression that the mechanisms that guide our actions instantaneously consider all the information that is available to our senses. However, our research on hitting moving targets suggests that this is not true. We found that there are many conditions in which subjects do not use the information that we would expect them to use. The clearest example is that subjects often failed to use the perceived velocity to anticipate where they will hit the target. This was so despite their having access to the information: the acceleration of the hand did depend on target velocity. Neither did subjects choose a position and adjust the acceleration of the hand so that they would reach that position at the same time as the target. They clearly relied on continuously adjusting their action. I will argue that this is a good strategy. In many cases it is more important to react quickly, than to react perfectly appropriately. Consequently our nervous system processes different aspects of information about our surrounding as simply (i.e. separately) as possible. Unfortunately, we are not yet able to quantify "simplicity" in this context, so we cannot directly predict what information should be used. What I intend to show is that there are cases in which potentially useful information is not used.
Set-size effects for spatial frequency change discrimination in multiple targets. Previous research has shown a substantial "set size effect" for spatial frequency change detection thresholds, suggesting a low-level explanation for "change blindness" (Wright et al 2000). Our stimuli consisted of two 150 msec frames, each containing 4 Gabor targets, and separated by a 250msec ISI. We measured spatial frequency discrimination of corresponding targets across the two frames. The set size was varied by pre-cueing 1, 2, 3 or 4 of the Gabors, and response competition was eliminated by post-cueing only one of the four stimuli. The spatial frequency of all stimuli was randomised, such that the task could not be solved by visual search in a single frame but required a comparison of two frames. In the first condition, all targets changed at random on any trial, but only the post-cued change was relevant to the discrimination. In a second condition, only one of the four targets changed on any trial, and the changing target was either post-cued or uncued. The log-log slope of Weber fraction with set size was 3-6 times steeper than reported for visual search. The similarity of results for "one change" and "all change" conditions suggests, in a signal detection analysis, that noise due to the changing stimuli is small in relation to internal noise, and averaging of the stimuli is not a factor.
Attentional modulation of visual adaptation: its spatial spread and We investigated the effects of diverting attention on the movement aftereffects (MAEs) produced by drifting gratings. MAEs were measured throughout their lifetimes, by matching their apparent velocity with a physically moving grating. When subjects named the sequence of rapidly changing digits at the fixation point, peak MAE velocity, as well as duration, was reduced. The reduction was greatest when the inner edge of the adapting grating was close to the fixation point, but was still large at a distance of 3 degrees. As found by others, MAEs decayed exponentially with time after adaptation. The effects of diversion were to reduce initial velocity and increase the rate of decay. This effect is more like that of a decrease in adapting duration than that of a decrease of adapting contrast or velocity, and sets constraints on theories which compare attention to a variation of 'signal strength'.
It used to move but now it's stuck: head centred motion perception in the Stationary objects appear to move opposite to a pursuit eye movement (Filehne illusion) and moving objects appear slower when pursued (Aubert-Fleischl phenomenon). Both illusions imply that extra-retinal, eye-velocity signals lead to slower estimates of speed than corresponding retinal motion signals. Intriguingly, the Filehne illusion has been shown to reverse at short but not long durations in older observers (Wertheim & Bekkering (1992) Vision Research, 32, 2379-2384). This suggests an interaction between age, duration and signal size. To test the signal-size hypothesis, we compared the strength of Filehne illusion and Aubert-Fleischl phenomenon in young (c. 20 years) and old (c. 70 years) observers and at short (200 ms) and long (700 ms) display durations. At long durations there was no evidence that the size of the Filehne illusion depended on age whereas at short durations there was. The age-dependent change was not quite as pronounced as that reported by Wertheim & Bekkering - instead of reversing in older observers the Filehne illusion disappeared on average - but the trend was in the reported direction. More remarkable was the fact that all observers showed similar Aubert-Fleischl phenomena regardless of age or duration. The differences between the two illusions could not be reconciled on the basis of actual eye movements made. Our findings therefore cast doubt on the signal-size hypothesis.
Effects of retinal image degradation and perceptual learning on INTRODUCTION
A Comparative Study of Potential of Generic and Custom-devised Image Introduction: Early studies have indicated that image enhancement holds potential for increasing the visibility of images for people with low vision. This study is new in comparing the effectiveness of a range of generic and custom-devised filters. Methods: The following filters were included : histogram equalisation, edge detector (Sobel), contrast stretch, difference of Gaussian, unsharp masking, Peli's adaptive thresholding and enhancement, anda custom multiplicative filter based on the subject's contrast sensitivity loss. Most of these filters may be applied with variable parameters. Within-filter comparison was by subject ranking and between-filter comparison was by subject rating. Twenty-eight subjects took part, aged 40-80 years, with low vision due to a variety of disorders. Visual characteristics (visual acuity, visual field loss, contrast sensitivity, and visual disorder) were studied for ability to predict filter preference. Results: The within-filter ranking showed that filters with high gains and exaggerated enhancements were not preferred by subjects and they were eliminated from further study. The between-filter ratings showed that the most frequently preferred filters overall were Peli's adaptive enhancement, unsharp masking and the contrast stretch. Peli's adaptive thresholding and the custom multiplicative filter produced a detrimental effect for most subjects. The only visual characteristic which was associated with filter preference was visual disorder.
Time Delays in Head-Mounted Displays The Variable Latency Asynchronous Display (VLAD) is a special-purpose experimental device that has been designed and built to study aspects of delay such as are encountered in visually-coupled systems (VCSs). It is a system consisting of cameras and a binocular display mounted on a helmet, together with a video delay unit, and it allows captured images to be presented at 120 Hz, with an intrinsic delay of less than 10 ms to which a further delay of up to 155 ms can be added in 5 ms increments. It has been used to examine the effect of delay on various aspects of visually-guided movements. In a first experiment it was used to demonstrate the consequence of these delays in a pursuit tracking task. In a second experiment, it was used to introduce delay into an experiment on displaced vision, but no adaptation to a lateral displacement of 11ƒ could be demonstrated for added delays of 0, 45 or 95 ms. It is thought that problems common to head-mounted systems, including helmet slippage, might have countered the adaptation effect. This has led to a re-examination of the accepted evidence concerning prism adaptation in more impoverished conditions and its abolition by delay. VLAD has permitted other informal observations related to time delays and is currently being used in an experiment on stereopsis. This work was funded by Technology Group 5 of the UK MoD's Corporate Research Programme.
Visual adaptation: Enhancements in signal discrimination, while fixing signal to noise ratio. Visual adaptation is a fundamental sensory process that describes the decrease in response of a sensory system to a sustained input signal. Current theories posit that visual adaptation allows visual neurons to respond with high sensitivity to a wide range of visual inputs in order to compensate for the limited dynamic range that is attributed to individual neuron transfer functions. Here, a further constraint is proposed to help explain properties of adaptation. It is suggested, that adaptation fixes the signal to noise ratio of neuron responses with the added benefit of enhancing signal discrimination. Drawing upon existing Psychophysical data that has studied effects of adaptation upon both contrast and speed perception, it is posited that at least two independent effects of adaptation can be identified. One, which is fast, divisive and isotropic in the spatial domain can be explained by adjustments in the temporal bandwidth of simple cell transfer functions. The mechanism increases temporal bandwidth as a function of increasing image contrast while keeping signal to noise constant. This type of adaptation predicts that signal discrimination may be enhanced by an amplification of any high temporal frequency components present in the image signal. The second is proposed to be slow, subtractive and direction/orientation tuned. It is posited that this latter mechanism shifts the transfer function of simple cells where the slope of contrast response function is maximum towards the contrast of the adapting signal. The subtractive adaptation is proposed to enhance discrimination by fixing the signal to noise ratio of quantization noise about the contrast of the adapting signal at a maximal value. To illustrate these ideas, a computational model of motion perception is proposed. The contrast response of the motion model is shown capable of explaining the temporal properties of spatial contrast vision (fast adaptation) as reported by Georgeson (Vision Research, 765-780, 1987) . The speed transfer function of the model is shown capable of explaining the effects of the mentioned adaptive processes on motion perception (both perceptual increases and decreases in speed) to explain effects of motion adaptation on motion perception (both fast and slow adaptation) as reported by Smith and Edgar (Vision Research, 253-265, 1994).
Posters Visual attention is allocated within a retinotopic framework in response to At ECVP we reported that exogenous shifts of visual attention were made to locations encoded within a retinotopic framework (Barrett et al., 2000). Mark F. Bradshaw, Simon J. Watt, Tanya J. Holden, Kathleen M. Elliott, Paul B. Hibbard Binocular disparity is often considered paramount in the control of natural prehension movements (e.g. Servos et al., 1992; Watt and Bradshaw, 2000). Despite this little is known about how children use this information in the control of the reach and the grasp nor how the ability to exploit such information improves through middle childhood. Here we examine the ability of two groups of 10 children (5-6 and 10-11 yrs) to reach for objects of different sizes placed at different distances under binocular and monocular viewing conditions. Children reached for and lifted solid rectangular objects placed along the midline in normal lighting conditions. They wore LCD goggles which controlled the stimulus presentation and viewing condition. A MacReflex Motion Analysis system was used to record the movements and a full range of kinematic and timing indices were determined off-line. It was found that for both age-groups, peak velocity was a linear function of an object's distance under both monocular and binocular conditions. There was no apparent difference between monocular and binocular conditions. Maximum grip aperture increased as object size increased, although the magnitude of this scaling was less under monocular conditions. In conclusion the elimination of binocular cues appears to affect the grip formation while leaving the transport component relatively unaffected. This was true for both age groups.
Binocular cues do not provide accurate depth information for the control To reach out and pick up an object, one needs to know its location, shape and size. This information is potentially available from binocular visual cues. However, the perception of three-dimensional space on the basis of this information is known to be distorted systematically (e.g. Johnston, 1991). Here we investigate whether these distortions are also apparent when we reach for disparity defined objects. Prehensile movements to both real and virtual objects, of three different sizes, placed at two different distances along a tabletop, were recorded using a Macreflex motion analysis system. For both real and disparity defined, virtual objects, peak grip aperture scaled with object size and peak wrist velocity scaled with object distance. When reaching for disparity defined objects, subjects progressively underreached as the object distance increased. Grip apertures for the virtual objects placed at the nearer distance were consistent with an overestimation of object size. These results suggest that the representation of three-dimensional space derived from binocular information in the control of prehension is subject to similar distortions to those revealed using psyhophysical paradigms.
Biases in stereoscopic shape constancy. Binocularly viewing a stationary object generates horizontal disparities compatible with a one-parameter family of possible 3-D shapes. The correct shape is easier to identify under rich viewing conditions when different perspective projections reach each eye. This study examined whether shape was perceived more accurately when a) the distance specified by vertical disparities was enhanced by increasing display size from 8° ('S') to 33° ('L') and b) textural shape information was improved by replacing random dots ('D') with appropriately compressed circles ('C'). Vertically-hinged (open-book) stereograms with dihedral angles between 35° and 169° were simulated by vergence and differential perspective to lie at distances from 30cm to 300cm. Subjects monocularly adjusted the relative orientation of two lines to match perceived dihedral angle. They also set the height of a vertical line to the size of a real, hand-held tennis ball. Settings were made and stimuli presented on monitors positioned at physical distances of 34cm. Spearman rank correlations between set and simulated dihedral angles decreased in the order LC(0.95)>SC(0.90)>LD(0.83)>SD(0.79). When disparity depicted the same shape, subjects perceived less depth at larger distances.
Working memory and the control of natural prehension movements. An often overlooked aspect of natural prehension is the importance of working memory processes which are responsible for the selection and maintenance of goal directed behaviours. Working memory is subdivided into a central executive and two slave systems: the phonological loop and visuo-spatial scratchpad. The latter system is further subdivided into visual and spatial component, which may, in principle, encode extrinsic and intrinsic properties of objects in the world (see Baddeley & Logie, 1999).
Visual factors in the prevalence of left neglect Hemi-spatial neglect is a disorder of attention following cortical damage. We sought to explore the possibility that asymmetries in the distribution of attention between the two hemifields might contribute to the prevalence of left neglect. We used a visual search task, and measured reaction times. We added noise masks selectively to stimuli in the left or right hemifield to simulate the effects of inattention. Four of nine observers had significant asymmetries in the effects of noise, for three of them the asymmetry was such that noise in the left hemifield selectively masked targets in this hemifield. Most of the remaining observers showed a trend in the same direction, making this asymmetry significant overall. Only one observer had a significant asymmetry in the opposite direction.
Going round in circles? Shape and the Ebbinghaus illusion. Previous explanations of the Ebbinghaus illusion have centred on two main ideas: an object-based, size-contrast hypothesis and an edge-based, contour-interaction hypothesis. We tested these theories by measuring the illusion with the central test element of one shape and the surrounding inducer elements of another shape. Four different shapes (circle, hexagon, triangle and irregular) were tested in all 16 possible combinations. With circular test figures there was a systematic relationship between the illusion's strength and the degree of
Height in the field and binocular cues support natural prehensile movements Although binocular cues are typically considered pre-eminent in the control of prehensile movements, accurate and efficient movements are still possible under monocular viewing. The aim of the present study was to examine the efficacy of (i) binocular information, and (ii) height in the field (HIF) information in the control of reaching and grasping when both cues were available in isolation. To do this subjects reached for and picked up 3 different objects at 3 distances under binocular and monocular viewing in an impoverished visual environment (complete darkness with self-illuminated objects and hand) in which objects were placed along the line of sight to eliminate HIF information. A MacReflex motion analysis system was used to analyse the kinematic parameters of subjects' movements. |
||||||||
| webdesign by ablen | The AVA is a registered charity (No: 1049146) |