The Visual Display Unit (VDU)

O f course the VD U to w hich the im ages are being sent m ust also be capable o f operating at such high refresh rates, and such displays have been developed in recent years. T he technology m ost w idely used for rendering high-resolution

anim ations in psychophysics experim ents is the Cathode Ray Tube. The ‘ray ’ is a stream o f electrons em itted from the cathode (negative electrode) o f an electronic valve. T he screen is itse lf part o f this valve, and is coated w ith p h o sphor dots that fluoresce w hen struck by electrons, em itting light at a frequency that is a characteristic o f the phosphor used. In this way colours can be produced by using several different phosphor-types em itting light at different w avelengths. T he m ore closely p acked the p h osphors are the g rea ter the resolution an im age can be shown at. T he light em itted from the phosphors is determ ined by the degree o f control that can be exercised over the voltage applied to them . As noted above, the am ount o f light em itted from a phosphor is proportional to the voltage applied to it via the electron gun. The signal that m odulates this voltage exists as a digital representation in the com puter. As such it has a resolution that is dictated by the length o f the binary w ord used for it. For exam ple an 8-bit digital w ord has 256 (2^) com binations, and so is able to represent a num ber in steps of A 16-bit w ord has 65536 (2^^) com binations and can represent a num ber in steps o f — !— . As there are discrete steps in the

^ ^ 65536 ^

representation, neither word length is able to accurately represent interm ediate num bers. H ow ever, a 16-bit w ord can m ore closely represent a given value because its steps are m uch sm aller, reducing the quantization error. It w ould be advantageous to control the lig h t em itted from the screen w ith as great a resolution as possible, but the word length used in m ost contem porary graphics cards is only 8 bits. This m eans that a display capable o f producing 0 to 100 candelas o f lum inance can only do so in steps o f or 0.39 candelas. This

256

m ay be in su fficien t for som e ex perim ents th at dem and fin e r control over lum inance values, such as contrast threshold m easurem ent. One im provem ent to this lim itation is possible if the experim enter is content to display greyscale im ages, rath er than colour. Pelli & Z hang (1991) pro p o sed a m ethod for com bining the output o f the three colour channels to increase the resolution o f

the voltage applied to the cathode o f the display. The m axim um voltage to each electron gun is the same, but when passed through a potential divider resistor netw ork can be m ade sm aller to any degree required. This potential divider, know n as the ‘V ideo A ttenuator’ is used to reduce the voltage on each channel by differing am ounts. This m eans that each channel has a different voltage step size for each change o f a bit in the digital w ord controlling it. The voltages are then sum m ed together to provide a single output that inherits an im proved voltage resolution, equivalent to having voltages under the control o f a 12-bit word. If a greyscale m onitor is used w ith this signal then a greyscale im age is the result. C onnecting this output voltage to ju st one o f the RGB (Red, Green, Blue) inputs o f a colour m onitor results in a m onochrom e im age, which can be an acceptable course given the low availability and high cost o f greyscale m onitors. To achieve a greyscale im age on a colour m onitor, we have added a splitter/am plifier circuit to the V ideo A ttenuator to m ultiply its single voltage output to three identical channels, which are then applied to each o f the RGB inputs o f the VDU.

Linearising th e ‘G am m a’ function. For m any psychophysics experim ents it is essential for there to be a linear relationship betw een the num bers used to define an image in the com puter and the lum inance that the display produces. H ow ever it is not norm al for a com puter system to provide this desirable feature, and so the experim enter has to ensure this linear relatio n sh ip is achieved through softw are m eans. T his process is know n as linearising the d isp la y ’s G am m a function. First, the existing relationship betw een the tw o is characterised by m easuring the output o f the display for a large range o f input values. The correspondence betw een n um ber and lum inance can then be fitted w ith a function that captures the relationship betw een them . Finally, any non-linearity found can be com pensated for by creating a lookup table through w hich the desired im age values are passed to generate the correct lum inance at the display. A ge and changing environm ental conditions cause drift in the specification o f electronic com ponents, so the lin e a risatio n p ro ced u re has to be rep eated periodically to m aintain accurate control over luminance.

2.2 Methods

The follow ing section details m ethods com m on to m any o f the experim ental chapters to follow. D epartures from these general m ethods are described in the relevant chapters.

F orced choice p rocedures w ere used in all experim ents, n orm ally a single decision b eing m ade betw een tw o sp atial or tem p o ral locations; the ‘tw o alternative, forced c h o ice’ or ‘2A F C ’ procedure. In this paradigm one location contains a signal, the other no signal, the task being to discrim inate correctly betw een them . The locations in question could be either spatial o r tem poral, depending on restrictions im posed by stim ulus design and the abilities o f the hardw are to deliver the stim uli. Each chapter details w hether locations w ere spatial or tem poral, and in the case o f chapter 2 the requirem ent for observers to m ake two judgem ents (2*2A FC ) rather than one is described m ore fully. Spatial or tem poral locations alw ays m aintained a constant relationship to each other (e.g. spatial separation, or inter stim ulus in terv al), b u t th eir ordering w as assigned at random to prevent selective attention by the observers.

Stim uli. W ith one exception (see C hapter 3, Experim ent 1) the stim uli used in the thesis w ere supra-threshold, i.e. w ell above lum inance contrast threshold, the operationally defined m inim um signal level required for a criterion level o f perform ance. In this context the signal location referred to earlier com prises a proportion o f dots sharing som e organising principle, for exam ple direction o f m otion, and a p ro p o rtio n o f dots w ith no o rg an isa tio n along th at sam e dim ension. The no-signal location contains only d isorganised dots, and the d iffic u lty o f d istin g u ish in g the lo ca tio n s is m an ip u la te d by a lte rin g the proportion o f organised to disorganised dots in the signal location. This is known as the m otion coherence paradigm , and is the m ethod used in m ost o f these experim ents to establish threshold levels o f perform ance. A m otion coherence threshold is the ratio o f organised dots in the signal anim ation to the total num ber o f dots in that anim ation, and is quoted in percentage terms.

Each location com prised an annular window, curtailed at the edges by a contrast m odulation. T he m odulation at each edge fo llow ed a co sin e p ro file that progressively reduced the contrast at the edges from m axim um to zero. This is term ed a ‘raised c o sin e ’ because o f the extended m axim um contrast central region flanked by sym m etrical cosine-shaped decay.

Each anim ation was also contrast-m odulated in tim e by a raised cosine w indow to preclude disruptive transients at the beginning and end o f an anim ation. A nim ations generally lasted 1000m s, w ith app ro x im ately 500m s elapsing betw een the observers response and the subsequent trial.

A black cross w as show n w hich subjects w ere required to fixate during each trial. In experim ents involving tw o spatial locations the cross w as placed h a lf w ay b e tw e en lo c a tio n s and sh o w n c o n tin u o u sly . In te m p o ra l in te rv al experim ents the cross was shown betw een interval presentations.

Levels o f the independent variables w ere selected using the adaptive Q uest procedure (W atson & Pelli, 1983), except in C hapters 3 and 6, w here the m ethod o f constant stimuli was used.

Responses. O bservers initiated a block o f trials using the alphanum eric keypad o f the com puter. If the subject perceived the left spatial location (or first tem poral interval) to contain the signal, button T ' was pressed. A lternatively, button '3' w ould be pressed if the signal w as p erceived in the right spatial location (or the second tem poral interval). A new trial w as initiated only after their response, and there was no tim e pressure to answer. If the subject needed to take a break button '2' could be pressed to repeat the trial, w ith a new value o f the independent variable being selected.

Feedback. Except w here subjective judgem ents w ere required (chapters 5 and 6) auditory feedback indicating correct and incorrect judgem ents w as given after the o b se rv e rs’ response had been recorded. N o feedback w as given w here subjective judgem ents were called for.

V iew ing distance was generally 57cm or a m ultiple thereof, except in chapters 5 and 6.

D ifference o f G aussian dots w ere used throughout, for the reasons detailed above, though their sizes varied between experim ents. A Difference o f Gaussian (D oG ) dot was form ed by sum m ing two G aussian distributions, one positive going the other negative. The standard deviation o f the negative going Gaussian was 1.5 tim es greater than the positive. The spatial frequency bandw idth o f the D o G ’s was 1.77 octaves.

D ot lifetime. D ot lifetim es were tightly constrained. A dot with a lifetim e o f two w ould initially appear at a random location, ju m p to a new location to produce an apparent m otion, then expire. A new d ot w ould then be positioned at a random location to m aintain the overall quantity o f dots, and hence dot density. Synchronised birth and death o f the dots w as m inim ised by assigning each a random ly chosen elapsed lifespan at birth.

C ontrast: The M ichelson contrast o f the dots was 0.667, and the background lum inance was 50 candelas/m , except in the experim ents o f C hapter 7. Stimuli were view ed binocularly in a darkened laboratory.

Equipment

W ith the exception o f Chapter 7 a Sony Trinitron M ultiscan 400PS display was used throughout, its size being: 35(w ) * 26(h) cm . A Form ac G A 12 graphics card was used to drive the display at a resolution o f 832*624 pixels giving 24 pixels per cm. The refresh rate was typically 75Hz. The display used in C hapter 7 was a Phillips B rightview m onochrom e CR T (see Chapter 7 for details).

2

The m ean lum inance o f the display was set to 50 cd/m , and w as m ade a linear function o f the digital control signals from the graphics card using a M inolta light m eter. Pseudo 12-bit resolution was achieved using the ‘V ideo A ttenuator’ o f Pelli & Zhang (1991). The three colour guns (Red, G reen and Blue, RG B) of the display w ere driven equally with a custom -built splitter-am plifier o f our own design to produce grey level images.

Stim uli w ere generated on an A pple M acintosh G4 450M H z personal com puter, using softw are routines from the V ideo Toolbox (Pelli, 1997). In C hapter 7 a 667M Hz A pple M acintosh G4 was employed.

Chapter 3

Direction Bandwidths of Rotation, Radial and Translation Motion Mechanisms

3.1 Abstract

The visual system is sensitive to m any dim ensions o f the visual array, such as colour, spatial frequency and direction o f m otion. N euronal m echanism s that are sensitive to a lim ited num ber o f dim ensions are term ed ‘a n aly sers’. A useful strategy to code a w ide range o f values along one dim ension w ould be to sub­ divide the range am ong several analysers w ithin the same class. Such a schem e has been show n to exist in the processing o f colour and spatial frequency, and the m easurem ent o f the tuning w idths o f com ponent analysers is now w ell understood. Sim ilar inv estig atio n s have b een carried out for m echanism s sensitive to m otion direction, but this w ork has been largely concerned w ith u n id ire c tio n a l m otion. C om plex p attern s o f m o tio n are p rev a len t in the experience o f hum ans, and there is m uch evidence to suggest the existence o f specialised detectors for rotation and radial patterns. It w as the purpose o f these experim ents to m easure the tuning w idths o f such m echanism s.

3.2 Introduction

F irst a note on term inology. This chapter concerns psychological m echanism s responsive to a range o f directions o f m otion. Previous literature on this subject m ixes term s in a seem ingly interchangeable way, when, in fact, quite different widths are being referred to. For clarity the follow ing term s used throughout are defined here. The ‘B andw idth’ o f a m echanism is its width at h alf the m axim um response am plitude o f the m echanism , i.e. ‘half-h eig h t.’ The ‘fu ll-w idth’ o f a m echanism is its width at its m inim um am plitude, i.e. the m ost extrem e extent at w hich the m echanism responds to a stim ulus. Som e authors have referred in original papers to a m echanism ’s bandw idth by quoting h alf its w idth preceded by sign sym bols. For exam ple, ±30 degrees refers to a w idth o f 60 degrees. To avoid confusion, the unsigned form will be used throughout.

Behavioural evidence has shown motion sensitive m echanism s to be restricted in the range o f directions to w hich they are responsive (Levinson & Sekuler, 1975; Sekuler & Levinson; 1974, Sekuler, Pantle & Levinson, 1976). This range o f m otions is referred to as the ‘direction b andw idth’ o f the m echanism and has been estim ated using a variety o f techniques. W hile researchers agree that the direction bandw idths o f m otion detectors are in general term s ‘broad’, seldom do two experim ental paradigm s yield the sam e quantitative estim ate. For exam ple, psychophysics experim ents have proposed bandw idths as large as 150 degrees (Ball, Sekuler & M acham er, 1983), and as little as 70 degrees (Raym ond, 1993). Electro-physiologists have identified that a range o f directional tuning exists in V I . F o r ex am ple, A lb rig h t (1984) found som e c e lls ’ b an d w id th s to be considerably less than 40 degrees, yet m any to be as large as 100 degrees and m ore (mean o f 68 degrees), with interm ediate values also being represented. F inally, m odellers o f m otion detection m echanism s find bandw idths o f 60 degrees as a plausible m inim um figure to account for hum an direction sensitivity (W atam anuick, Sekuler & W illiam s, 1989; W illiam s, Tw eten, & Sekuler, 1991). Early experim ents by S ekuler and colleagues attem pted to characterize the directional sensitivity o f m otion sensors in term s o f the independence betw een

m echanism s. Several techniques were used. For exam ple, Levinson & Sekuler (1976) adapted subjects to a translating stim ulus o f (supra-threshold) random dots, then m easured the perceived direction o f m otion o f a subsequent test stim ulus o f sim ilar design. They found that the adaptation stim ulus significantly altered the perceived direction o f m otion o f the test stim ulus as long as one was within 90 degrees o f the other. This finding has been interpreted to imply that independence betw een detectors is not achieved at less than 130 degrees o f separation, (Ball, Sekuler & M acham er, 1983).

Ball & Sekuler (1979) also used supra-threshold random dot stimuli, this tim e in a forw ard m asking ex p erim en t. A noise m ask co m p risin g m otion in all