|
Promoting vision research and its applications |
|
||||||||
|
AVA 2002 Achromatic Vision 27 March 2002
AVA2002 Programme
10:00 Registration and coffee Posters M. Wright and T. Ledgeway AVA2002 abstracts Geoffrey J. Burton Memorial Lecture
Papers Second-stage Mechanisms: a Computational Model T.J. Atherton1, S.J. Hinds1, and K. Langley2, 1)Computer Science, University of Warwick, Coventry, CV4 7AL, and 2)Psychology, UCL, Gower Street, London. Email: tja@dcs.warwick.ac.uk We outline a model of second stage spatial processing in mammalian vision. The model is an extension of the familiar “filter-rectify-filter” scheme having the form of two functionally identical stages in series; Stage 1: “filter-rectify-pooling”, Stage 2: “filter-rectify-pooling”. We explicitly define the form of the “pooling” which is over orientation. The model has some interesting properties. The first stage produces multiple intermediate image maps that carry information about the presence of energy-orientation symmetries (zero-fold indicates energy within the filter’s pass-band, two-fold is edges or bars, etc). Also available as outputs of the first stage are phase-orientations that are derived by omitting the “rectify” from the processing flow. We do not consider phase-orientations in this paper, instead see Atherton (2002). We focus on the “demodulated” energy-orientations from the first-stage and how these might be further processed to provide insight into Contrast Modulated (CM) gratings and plaids, Orientation-defined texture edges Nothdurft, and the “Collator” units described by Moulden (1994). There are remarkable similarities between the first and second stages, but also differences. The first stage takes in a signal proportional to luminance and produces complex “feature maps” that carry information about energy-orientation. At each position the result is (literally) a complex number of the form a(x,y) + i.b(x,y) (or equivalently r(x,y).[cos(x,y) + i.sin(x,y)]), the map codes the presence of an oriented image feature as the magnitude of response, r(x,y), and the orientation of the feature as the phase, (x,y). The type of feature at each position is coded by which map the response occurs in. The second stage takes its input from the first-stage responses, and here is one difference, the input is a complex image. The second-stage filters are tuned to frequencies between two and four octaves below those of the first-stage filters. The processing at the second-stage takes the same form as at the first-stage. The complex input to the second-stage has the remarkable property that an orientation-defined edge is most easily detected by the second-stage when the energy-orientation from the first-stage output differs by 180 degrees across the edge. The talk will take a non-theoretical approach, outline the model, and be illustrated by examples of second-order stimuli processed by the model. Extensions of the model to include pooling over spatial frequency and its relationship to responses beyond the Classical Receptive Field will be discussed.
Subitization and attentional engagement by transient stimuli Louise Alston and Glyn W. Humphreys. Department of Human Sciences, Brunel University Uxbridge, Middlesex. Email: Louise.Alston@brunel.ac.uk A series of experiments were carried on the selection of moving and static items during enumeration. Small numbers of targets can be enumerated with little increase in reaction time and error, a process referred to as subitization (Kaufman, Lord, Reese and Volkmann, 1949). For moving targets among static distracters, there was efficient selection characteristic of subitization. This was not the case for static targets among moving or transient distracters, which showed much larger RTs and a steady increase in RT with target number. Efficient selection of the static targets was possible, however, when static targets and moving distracters were presented more foveally. With small inter-item spacing, grouping by proximity among the static items may have facilitated segmentation. This was tested in Experiment 2 by increasing the size of the static items in the more dispersed displays to increase grouping. Although reaction times were slightly improved, they still increased linearly with target number. A second possibility is that when targets fall across relatively wide spatial areas, subitization will depend on cells with large receptive fields if those targets are to be detected in parallel. Such cells also tend to be activated strongly by motion (Baizer, Ungerleider, and Desimone, 1991). In particular, cells in Area MT respond selectively to movement and have been suggested as the cause of a similar asymmetry between moving and static targets in visual search (McLeod et al., 1991). In contrast, with foveal presentation the receptive fields would be much smaller leading to a different response to moving stimuli. In these displays, the moving items were perceived as having a higher velocity than when the display area was large and seemed to form a coherent moving background. This increased segmentation of moving and static items appeared to be supporting parallel selection of the static targets. In an experiment manipulating the speed and spacing of the items, it was shown that when distractors moved more slowly, subitization of static targets among moving distracters broke down and there was no evidence for rapid enumeration of up to 4 targets. Ivry and Cohen (1992) suggested that cells responding to static stimuli are also affected by moving stimuli, when the velocity is relatively slow and large receptive fields are activated - processing may be biased toward the magno / dorsal system. Hence parallel detection of static items amongst moving ones may be difficult.
Summing Pictorial Depth Cues to Increment Threshold: A Similar Result to Binocular Disparity and Motion Tim S. Meese & David J. Holmes. Neurosciences Research Institute, Aston University, Birmingham, B4 7ET, UK. Email: t.s.meese@aston.ac.uk As any undergraduate student of visual perception knows, there are many routes by which the missing dimension of depth can be recovered from our two retinal images. Over recent years investigators have begun to ask how information from these different routes are combined to deliver coherent unified perceptions. While some experiments have inquired about combination across very different depth cues, including static pictorial cues and binocular disparity (e.g. Johnston et al, 1993, Vision Research, 33, 813-826), these investigations have effectively adopted matching or nulling techniques. Consequently, it is unclear whether these experiments reveal properties of sensory mechanisms or cognitive strategies. In a more incisive approach, Bradshaw and Rogers (1996, Vision Research, 36, 3457-3468) have gathered performance data using psychophysical probes of adaptation and summation-to-threshold and concluded that vision contains neural mechanisms for summing depth cues, at least from sources of motion parallax and binocular disparity. Here, we extend the use of the summation technique to investigate the combination of three static achromatic depth cues using stimulus elements arranged in nine rows and nine columns (see Fig 1). Fig 1 Stimuli were viewed monocularly through a black viewing tunnel to enhance perception of depth. Stimulus duration was 100 ms and the width of the middle row of elements was 10.2 deg. The null stimulus, which carried each of three independent depth cues plus a gentle compression of the distance between the rows, is shown in Fig 1A. Test stimuli were similar, except that they contained increments (or decrements) in one or two of three cues (gradients in: element size, row width and element contrast are shown in Fig1 B, C and D respectively). In each session, practised observers had to discriminate a test stimulus from the null stimulus in a two interval forced-choice design. Cue-increment thresholds were measured for (i) each of two different cues and then (ii) a compound stimulus containing normalised levels of the two cues. This whole procedure was repeated at least five times for each pair of cues. To remove static positional cues of individual elements, the stimulus was jittered by up to ±0.5 of the width and height of the central element on each presentation. For both observers, substantial summation was found between size and contrast cues (B & D), but only when both cues were increments; when one of the cues was a decrement, summation was abolished. One of the observers also showed substantial summation for the other two pairings: (1) width and contrast and (2) width and size. In a control condition, neither observer showed summation for any of the three depth cues with an orientation-gradient that was included in both the null and test intervals (not shown). We conclude that human vision contains summing mechanisms with inputs from at least some of the static pictorial depth cues. Whether this is the same or a different mechanism from that reported by Bradshaw and Rogers (1996) remains to be investigated. This work was supported by a pilot-study grant awarded by the Leverhulme Trust
A Fourier model for predicting letter acuity in peripheral vision. Roger S. Anderson1 and Larry N. Thibos2, A major goal of vision research is to develop computational models that predict visual performance on various tasks. One such task is letter acuity, for which a variety of models have been developed for foveal vision (e.g. Parish D.H. and Sperling G., Vision Research, 1991; Alexander K.R. et al, JOSA A, 1994; Solomon J.A. and Pelli D.G., Nature 1994). However, models of letter acuity for peripheral vision are underdeveloped by comparison. Two possible reasons for this state of affairs are that (1) peripheral vision has not been studied as extensively as foveal vision, and (2) the different factors which limit peripheral resolution (neural undersampling and aliasing) preclude a simple importation of foveal models into peripheral vision. To develop a model of peripheral letter discrimination, we measured threshold size for discriminating 33 pairs of Snellen letters at 30 degrees eccentricity in the nasal retina for two subjects, after correction of optical defocus. These pairs were chosen so that they differed in power by less than 10%. As a measure of psychophysical dissimilarity (D) for each pair we subtracted their images in the spatial domain, performed a Fast Fourier Transform on this difference image, and divided the RMS power in the resultant 'difference spectrum' by the sum of the RMS powers of the individual letter spectra. A plot of D vs. psychophysical threshold letter size gave a correlation of R=0.8. Threshold letter size varied by a factor of 4 depending on which letter pair was being discriminated. When D was calculated for letters that were low-pass filtered at different cutoff frequencies ranging from 2.5 down to 0.625 cycles/letter (c/let), the correlation with psychophysical performance was greatest when cutoff was 1.9 c/let (R=0.85) and only declined significantly when cutoff fell below 1.25 c/let. Conversely, when the difference spectrum was high-pass filtered at different cutoff frequencies, the correlation decreased continuously as the cutoff increased. These results imply that the band of frequencies between zero and 1.9 cycles/letter gives the best prediction of psychophysical discriminability of Snellen letters in peripheral vision, and is in close agreement with previous studies of foveal letter acuity indicating that spatial frequencies below 2 c/let are most useful for letter discrimination.
Looking at op art from a computational viewpoint. Johannes M. Zanker Department of Psychology, Royal Holloway, University of London Egham, Surrey TW20 0EX, England. Arts history tells an exciting story about repeated attempts to represent features, which are crucial for understanding of and acting in our environment, and which at the same time go beyond the inherently two-dimensional nature of a flat painting surface: depth and motion. In the twentieth century, marked by the invention and development of kinematic media (cinema, television, internet), some artists such as Bridget Riley began to experiment with simple black and white patterns that can create vivid dynamic illusions in static pictures. The cause of motion illusions in Op Art paintings is yet a matter of debate, but there are good indications that eye movements may play an important role in this phenomenon. In order to gain some insight on the possible consequences of retinal image shifts, synthetic wave gratings, dubbed as 'riloids', were used as basis for a geometric analysis and as inputs for a computational model of early motion processing. This two-dimensional array of motion detectors (2DMD model) provides response maps that represent the spatial distribution of motion signals generated by a stimulus, such as a two-frame sequence reflecting a saccadic displacement. The (moiré) interference pattern of the difference between the pattern at two stimulus locations can be interpreted luminance change necessary to detect local motion. Based on this geometry one would expect regions of motion in different directions, which is confirmed by first simulations providing motion signal maps in which local directions form extended patches of similarity. The direction usually does not correspond to the direction of pattern displacement, which can be expected as an instance of the so-called 'aperture problem' from the geometry of the curved gratings, giving rise to a variety of phase shifts and orthogonal motion components in the different regions of the pattern. The patchy structure of the simulated motion detector response to the displacement of riloids resembles the motion illusion, which is not perceived as coherent shift of the whole pattern, but as a wobbling and jazzing of ill-defined regions. Although other explanations are not excluded, this might suggest that the puzzle of Op Art motion illusions could potentially have an almost trivial solution in terms of small involuntary eye movement leading to image shifts that are picked up by well-known motion detectors in the early visual system.
Direction repulsion: Evidence for speed tuned velocity adaptation and a principal axis bias. Keith Langley1 & Stephen J Anderson2. Background: The perceived direction of a moving test sinusoidal grating may be altered by adaptation to another moving grating whose direction of motion is different. It is suggested that the adaptive changes in perceived direction, which are generally repulsive, are controlled by two main factors: (i) speed tuned velocity adaptation; and (ii) a principal axis bias. Methods: The test and adapting stimuli were sinusoidal gratings of 3.5 cpd periodicity and 8.0 Hz drift rate. To measure the direction of motion of the test patterns, subjects were required to determine whether the spatial orientation of a line marker appeared more clockwise than the direction of motion of the test patterns. Both the absolute and relative direction of the adapting and test gratings, and their contrast, were varied to determine the effects of motion adaptation on perceived direction. Results: When varying the absolute orientation of the test and adapting grating, but fixing the direction difference between them at 45o, direction repulsion was maximum when the adapting grating direction lay along the principal axes. Direction repulsion was minimum when the adapting grating moved along an oblique direction. When varying the relative direction between the test and adapting grating for two different adapting grating directions, the maximum shift in the direction repulsion in both cases occurred when the test grating moved in the direction of a principal axis. When varying the relative contrast but fixing the direction of the adapting and test gratings, the direction repulsion was independent on the contrast of the test signal for a high contrast adaptor. Conclusions: The results suggest that the perceived shifts in direction repulsion are a consequence of a speed tuned adaptive mechanism. The results are consistent with the predictions made by a quadratic programming model of motion perception in which both speed and direction are coded by opponent mechanisms that lie along the principal axes.
Posters Interaction between luminance gratings and disparity gratings Michael Wright1 and Tim Ledgeway2, 1Department of Human Sciences, Brunel University, Uxbridge, Middlesex UB8 3PH, U.K. and 2School of Psychology, University of Nottingham, University Park, Nottingham, NG7 2RD, U.K. Email: michael.wright@brunel.ac.uk Lowered stereoscopic thresholds were found in some random dot disparity gratings of spatial frequency f, when luminance gratings of spatial frequency f or 2f were added in the same orientation. The facilitation depended on relative spatial phase. However, when the phase relationship between the luminance and disparity gratings was manipulated such that relative phase was unpredictable, the threshold facilitation was eliminated. Thus, the lowering of threshold can be attributed to the spatial cueing effects of the luminance grating and the removal of spatial uncertainty. This was confirmed by showing that the presence of an explicit spatial cue, in the form of extended horizontal lines that differed from the remainder of the display only in terms of the granularity of the random noise texture, produced maximum facilitation. In a further experiment it was shown that a luminance grating of frequency f added to a suprathreshold disparity grating of frequency f or 2f was seen by the majority of observers as a shading pattern on a 3D corrugated surface. Depending on spatial phase, the luminance grating can modify the perceived shape of the stereoscopic corrugated surface. The possibility of suprathreshold interactions is suggested by biases in judgements of apparent depth.
Observers exploit binocular disparity information in motor tasks within dynamic telepresence environments. M.F. Bradshaw, K.M. Elliott & S.J. Watt. Department of Psychology, University of Surrey, Guildford, UK. Email: m.bradshaw@surrey.ac.uk Binocular disparity has, become increasingly commonplace in telepresence systems despite the additional cost of its provision. Experiments which show that perceptual and visuo-motor performance is worse under monocular than binocular viewing are often cited as justification for its use. Here we question whether this experimental comparison and provide an important set of data which compares performance on a motor task under binocular, monocular and bi-ocular (where both eyes receive the same view) conditions. Binocular cues were found to be particularly important in the control of the transport component. In the binocular conditions peak velocity scaling with object distance was greater than in the other conditions, and in the bi-ocular condition, where the binocular distance cues conflicted with pictorial information, no scaling was evident. For the grasp component, even in the presence of conflicting size and depth information, grip scaling remained equivalent in all conditions. For the transport component at least, binocular cues appear important and the decrease in performance observed in behavioural studies under monocular conditions is not attributable to lack of information in one eye but rather to the lack of binocular depth cues. Therefore in the design of telepresence systems to be used in telemanipulation tasks, the use of stereoscopic display technology seems justified.
Effects of Visibility and Separation upon Vernier Acuity for Narrowband and Broadband Stimuli. Brendan T. Barrett, David Whitaker & Paul Morrill. Department of Optometry, University of Bradford, Richmond Road, Bradford, BD7 1DP. Email: B.T.Barrett@Bradford.ac.uk Purpose: This study investigates the influence of visibility on vernier acuity thresholds for abutting and separated narrowband stimuli (Experiment 1), and asks whether these data can account for broadband vernier performance (Experiment 2). Methods: Experiment 1: Vernier thresholds were determined for sinusoidal grating stimuli at two spatial frequencies (1 & 8 c/deg.) across a range of contrasts (0.05-0.8) and exposure durations (35-2100msecs). Performance was assessed for the abutting configuration, and when a gap equivalent to 0.5 to 2.5 times the spatial period of the grating was introduced between the upper and lower halves of the grating. Experiment 2: Vernier thresholds were determined for a square-wave stimulus as a function of contrast (0.06 to 0.78). Exposure duration was fixed at 2100msecs. Thresholds were determined at the appropriate contrast levels for the fundamental frequency (1.8c/deg.) of the square-wave, and for a number of the harmonics (3F, 5F, 7F, 9F). Results: Experiment 1: Abutting condition: As expected, vernier performance was found to improve as a function of contrast at all exposure durations for the lower spatial frequency (1c/deg). This was also true for the higher frequency (8c/deg) at short exposure durations. However, performance exhibited considerable contrast independence when longer durations were employed. Separated conditions: As the separation between the target elements was increased, vernier performance became increasingly independent of contrast. This trend was more evident at the lower frequency. Experiment 2: Abutting condition: At high contrasts of the square-wave, vernier sensitivity is roughly equal for the F, 3F, 5F & 7F components. However, square-wave vernier performance is consistently better than the best performance for the component frequencies. Separated condition: When a gap equivalent to 0.5 times the spatial period of the fundamental frequency is introduced between the stimulus elements, it appears that only the fundamental and third-harmonic contribute to square-wave vernier performance. Conclusion: Vernier performance for abutting stimuli can be invariant across much of the contrast range, provided high frequency stimuli and long exposure durations are employed. Despite this, contrast independence is not exhibited for abutting broadband stimuli because, within the broadband stimulus, the contrast of the higher harmonic components never reaches a level which is high enough to reveal this plateau. As with detection of broadband stimuli, we suggest that vernier performance for abutting and closely separated broadband stimuli represents the envelope of vernier sensitivity of those spatial frequency mechanisms activated by the broadband stimulus.
Do depth aftereffects arise pre-attentively? David Rose1, Mark F Bradshaw1, Paul B Hibbard2. 1Department of Psychology, University of Surrey, Guildford, Surrey GU2 7XH and 2Department of Psychology, University of St Andrews, St Andrews, Fife KY16 9AJ. The computation of depth from binocular disparity is normally considered to occur in early areas of visual cortex. In contrast, the computation of motion may take place in early or in higher areas (V5/MT). Following the work of Chaudhuri (Nature, 344, 60-62, 1990) the mechanisms of motion perception have been established by several workers as vulnerable to the effects of attention, in that the motion after-effect is reduced if attention is focussed on a character-processing task during the adaptation phase. Here, we extend this method to the depth after-effect, using random-dot stereograms comparable in layout to the stimuli used for assessing motion after-effects. Thus for the depth after-effect, subjects adapted to stationary random-dot patterns which formed two rectangular patches standing out by unequal amounts in front of a stationary background plane, one rectangle above and one below fixation. Aftereffect duration was tested on stationary test patterns containing two rectangles equally prominent in depth. Similarly, for motion after-effects, we adapted with two rectangles of random dots drifting in opposite directions (left and right) above and below fixation against a stationary backgound. Testing was then performed with the same stationary test stimulus as for the depth after-effect. In some conditions subjects detected, during adaptation, target letters in a stream of characters, which were presented at one of two rates. We found that the durations of both after-effects were reduced in the conditions in which the character stream was processed during the adaptation phase. Moreover, the amount of the reduction increased in both tasks as the character stream became more rapid. The results are discussed in the light of recent theories that attention can modulate the earliest stages of visual processing, and it is suggested this must apply to depth as well as motion perception.
Exhibitors
|
||||||||
| webdesign by ablen | The AVA is a registered charity (No: 1049146) |
