perception in

perception in graphic displays

Roger A. Browsea,b, James C. Rodgera, and Robert A. Adderleya

b

Department of Computing and Information Science,

Department of Psychology, Queen’s University, Kingston, Ontario, Canada

ABSTRACT

a

Effective computer graphic applications should accurately convey three dimensional shape. Previously, we investigatedthe contributions of shading and contour, specular highlights, and light source direction to three dimensional shapeperception. Our experiments use displays of convex solid objects based on the superquadric parameterization,permitting continuous variation in their cross-sectional shapes. Our present work concerns the impact of surface

markings. Rotating wireframe or uniformly shaded objects may produce perceptually distorting shapes. We investigatethe idea that such distortions interfere with shape judgements, and that surface markings may either enhance perceptualaccuracy by encouraging stability, or impair it by interfering with global shading patterns. Our displays include rotatingobjects with no surface markings, stripes, latitudinal or longitudinal stripes, each with two different scene illuminations. Observers view pairs of objects, a target shape and a second object whose shape they adjust, using mouse clicks, tomatch that of the target. Our principal result is that these surface patterns do not enhance performance, even though thechosen stripe intensities minimise interference with global shading, and the stripe patterns may actually encode surfacecurvature. We are now investigating alternatives for applying surface patterns to modelled objects, including hardwaresupported texture mapping. Our long term goal remains the identification of a comprehensive set of conditions foroptimising shape understanding of graphic objects.Keywords: shape perception, surface patterns, 3D graphics

1. INTRODUCTION

The effectiveness of computer graphics applications such as scientific visualization, telerobotic interfaces, and

computer aided design depends on accurately portraying the three dimensional shape of objects. Generally, shapeunderstanding can be enhanced through the use of graphic display techniques that render the surfaces more accuratelyand realistically. However, our previous research showed that certain of the graphic rendering variables are moreimportant than others when observers attempt to judge the shape of displayed curved-surface objects. We investigated

the relative contributions of shading and occluding contour, the lack of contribution of specular highlights, and the effect

1,2

of light source direction on the accuracy of perceiving three-dimensional shape.Our techniques for examining shape perception use the display of three-dimensional solid shapes based on

3

superquadric parameterization. Variations in the superquadric parameters produce objects varying continuously in theshape of their orthogonal cross-sections. We use the range from a sphere to a cube to present a wide variety of convexobjects. Each has a simple numerical representation of its shape that we can use to measure performance in shape

discrimination and matching tasks. The displayed objects rotate about two axes to encourage observers to apprehend thecomplete shape rather than forming judgements on the basis of occluding contours. In our present research, we use thesame approach to address the issue of the effects of surface markings on the perception of object shape.

2. SURFACE MARKINGS IN THE PERCEPTION OF SHAPE

When the display of a rotating solid has the shading due to lighting removed so that only the occluding contour

is presented, observers often report the perception of a stationary but distorting object. In our previous work withshaded, rotating objects of uniform reflectance, observers still sometimes reported noticing distortion of the rotatingobject. We believe that these conscious experiences of distorting shape contribute to errors in shape judgement, andfurther, that subliminal distortions may interfere with the judgement of shape. It is a reasonable conjecture that thepresence of surface markings on a rotating object will encourage the perception of rigidity and thereby enhance shape

perception accuracy. Another way to view this issue is to consider the idea that accurate perception of the shape of arotating object may rely on an understanding of the movement of the individual locations on the surface of the object. Iftrue, then it seems reasonable to expect that the application of surface markings would enhance the observer’s stableidentification of surface locations, and hence enhance shape perception.

On the other hand, surface markings may reduce accuracy of shape perception. In our previous studies we

have shown that the presence of shading across the surface of an object is a significant source of information in shape

1

understanding. The presence of surface markings may impede the observer’s integration of these shading patterns

across the surface of the object resulting in less accurate shape perception. Certainly there is a convincing argument thatin the extreme, as in the case of camouflage, surface marking may hinder and distort perception.

There is a wide range of possible surface markings that could be tested for their influence on the perception of

object shape. As a first step in understanding the role of surface markings in shape perception, we have used a simpleform of striping applied to the superquadric shapes. The objects used in the unstriped condition have a brightness that isthe average of the two brightnesses that alternate in the striped objects, thereby controlling for overall brightness. Thebrightness levels used in the striping were selected to have a low contrast in order to minimally interfere with theshading across the surface. We used two different directions of illumination.

3. EXPERIMENTAL PROCEDURE

We generated the displays for the experiment using a Silicon Graphics O2 system with a 19 inch monitor. The

interface for the experiment, illustrated in Figure 1, was a straightforward extension of that employed in our previousresearchm. In that research, observers adjusted a two-dimensional contour to attempt to match the cross-sectional shapeof a rotating superquadric object displayed alongside it. In the current experiment, the observers viewed a pair ofsuperquadric objects in the Animation Window, and adjusted the right hand (adjustment) object to match the shape ofthe left hand (target) object. They made the adjustments by clicking with the mouse on either the circle or square in theAdjustment Window, located immediately below the adjustment object.

We varied conditions relating to

both the objects and the renderingparameters. Since our primary interestconcerned possible contributions of surfacepatterning to shape perception, the pairs ofobjects in each display could be in one ofthree surface patterning conditions. Thesewere either latitudinal stripes, longitudinalstripes, or uniform surface (no striping). Weproduced the stripes by varying the materialproperties of strips of the quadrilateralprimitives used to tessellate the surfaces. While the stripes for the two objects in a pairwere in the same direction, both the width ofthe stripes and the alternation of light anddark differed, preventing observers fromsimply matching stripes to achieve a globalshape match. Figure 2 depicts the threedisplay conditions applied to differentsuperquadric shapes.

The superquadric parameterization

of shape allows for the definition of cross-sections that range from square (at

ε=0.0) to circular ( at ε=1.0). Each

object is defined by two cross sectionparameters (ε1 and ε2 ), and the object’sextension along each of the local Cartesianaxes. The shapes that we used all had anaspect ratio of 1:1 in the horizontal plane,and had an aspect ratio of 1:1.4 for each ofthose axes relative to the vertical axis. Thecross-sectional shape of the target objectranged from 0.2 to 0.8 in eight equalincrements of the superquadric shape

parameters, with ε1=ε2. The shape of theadjustment object always differed initially,

offset by either 0.135 or 0.18 from that of the target object, and equally often towards either the square or round end ofthe shape parameter scale.

When we found, in preliminary data, that observers were exhibiting a bias in their shape matches, tending to

make the squarer objects rounder than the exact match, we decided to add another condition to the experiment. Weconducted the initial testing using a light source direction 45 degrees to the left and elevated 70 degrees. For the fullexperiment, we added a second light direction condition, with the light source direction coincident with the viewingdirection (that is, from directly in front of the objects, with no elevation).

These variations produced 108 distinct combinations of conditions, in addition to the two light source

directions. Each observer participated in one light direction condition, and viewed two replications of each of the othercombinations of variables, for a total of 216 trials. We recruited our participants by E-mail postings to the graduate andundergraduate students in the Department of Computing and Information Science at Queen’s University. There were 12observers for each of the light directions, for a total of 24 participants in all. We used a different random ordering oftrials for each observer.

During a trial, both of the objects rotated continuously with a compound motion. This combined complete

rotation about the vertical axis with swaying back and forth through an arc of 60 degrees about a horizontal axis. Thestarting points for the swaying motion differed between the two objects, so that, at any given time, the observer did not

see matching views of them. The observer clicked repeatedly on the shapes in the Adjustment Window, until a

satisfactory match was achieved, and then initiated the next trial by clicking on the Test Window. There was an upperlimit of 30 seconds per trial, so that if the observer had not clicked on the Test Window, the system automaticallyinitiated the next trial.

For each trial the testing software recorded the final shape of the Adjustment object, the elapsed time and the

number of clicks the observer used to arrive at the match. We analysed these data using the SYSTAT for Windowsstatistical package.

4. EXPERIMENTAL RESULTS AND DISCUSSION

Figure 3 shows the mean error across all 24 observers for each of the three striping conditions. An analysis of

variance shows that this a reliable difference in error for the different surface patterns. This favors the interpretation thatthe striping has interfered with, or camouflaged the shading across the surface and reduced the informative aspect of thatcue to a greater degree than the presence of striping has increased the shading through stabilization of the objects.Though not statistically significant, the mean error for the longitudinal striping is greater than that of the latitudinalstriping. If the striping were stabilizing the object under rotation, it is reasonable to expect the opposite effect. Therotation of the object is about the vertical axis, and so a perceived distortion of the horizontal cross-section would bedifficult when the object is striped longitudinally, because to perceive such a distortion would require an alteration in theperceived width of the striping. The latitudinal striping would not introduce that requirement.

Surface Pattern: Absolute Error [24 subjects]0.330.320.310.30.290.280.270.260.25latitudelongitudenonesurface pattern typemean signed errorShape: Signed Error [24 subjects]0.020.0150.010.0050-0.0050.2750.4250.5750.7250.350.650.20.50.8mean errorsuperquadric shape parameter Figure 3. Mean error for each surface pattern. Figure 4. Signed error for each shape value.

The main conclusion of our experiment is that forms of striping which one would expect to enhance

performance, in fact degrade the accuracy of shape perception. This is true in some instances for striping carefullychosen with low contrast to minimally interfere with the visibility of the surface shading due to lighting. We found in

some cases, that the accuracy of perception was degraded with striping which, in the spacing of the stripes, actuallyencoded the surface curvature. Thus there may be a surprisingly small range of surface striping that will enhance theaccuracy of shape perception for arbitrarily rotating and illuminated objects. For rendering and rotation configurationsthat induce object distortions, it may be possible to devise striping perpendicular to the direction of the distortion, thusoffsetting the distortion with bands that cannot be seen to change in width.

A second main result is the perplexing fact that observers tend to systematically overestimate the cross-sectional superquadric parameter. Figure 4 shows the mean signed error across all subjects and conditions plottedagainst the target superquadric shape parameter. We expected there to be comparable error levels for underestimationand overestimation of the parameter because the shape that was being adjusted began equally often with offsets on eitherside. A significant component of the error resulted from situations in which the adjustment object was initially moresquare than the target, with the observer adjusting past the point where the two shapes were identical, and beyond to thepoint of responding with an overestimation of the roundness of the cross section of the object. The graph in Figure 4 alsoindicates the statistically significant result that this bias towards overestimation increases as targets become more squarein cross section

We first noticed the bias towards overestimation when observers were carry out the task with a scene lighting

direction that was always above and off to one side. In order to find out if this lighting was contributing to the bias, wehad a second set of 12 observers carry out the task with scene lighting that originated down the viewing axis. The biaswas also present for this second set of subjects, ruling out lighting as the cause. In fact, light source direction had nosignificant effect on the estimation of the superquadric parameter, and so the results shown in Figures 3 and 4 representboth of the lighting conditions combined.

In the course of this ongoing research program we have developed a variety of techniques to measure human

performance in the judgement of shape. These techniques include forced choice similarity, same-different judgements,cross sectional matching, and now shape matching. We have used these techniques to examine a wide variety of

parameters that control the presentation and rendering of graphic objects, of which local surface reflectance properties isour latest concern. Our long term goal is the identification of a complete set of presentation and rendering conditionsthat optimize the accuracy of human judgement of the shape of graphic objects.

ACKNOWLEDGEMENTS

The research reported in this paper was conducted with financial support from the Institute for Robotics and IntelligentSystems and the Natural Science and Engineering Research Council of Canada.

REFERENCES

1.2.3.

R.A. Browse and J.C. Rodger, \"Investigations of three-dimensional shape perception for telepresence usingsuperquadric primitives\ Proceedings of the SPIE, Volume 2179, Human Vision and Electronic Imaging I,pp.235-246, 1994.

R.A. Browse, J.C. Rodger, and R.A. Adderley, \"The effect of lighting direction on the perception of shape ingraphic displays\Proceedings of the SPIE, Volume 2179, Human Vision and Electronic Imaging II pp.353-359,1997.

A.H. Barr, \"Superquadrics and angle-preserving transformations\pp.11-23, 1981.