Published in: Optics and Imaging in the Information Age, 42-50, 1997.

The Effect of Perceived Transparency on

Brightness Judgments

Mark E. McCourt1, Barbara Blakeslee1 and Frederick A. A. Kingdom2

1Department of Psychology 2McGill Vision Research Unit

North Dakota State University 687 Pine Av. W. Rm. H4-14 Fargo, ND, 58105-5075 Montreal, Quebec, H3A 1A1

USA Canada

Abstract Subjects matched the brightness of test patches located within a larger surround, where the surround was made to appear either different in reflectance from neighboring regions, or of the same reflectance but viewed beneath a transparent film. In both conditions the luminance and spatial extent of the immediate surround was equivalent, thus controlling for the effects of local contrast. Perceived transparency had a small but significant effect on brightness. Test patch brightness was most significantly elevated when the perception of transparency was supported by stereo depth cues. The effect was, however, mediated by the virtual transmittance of the transparent overlay, increasing in magnitude with decreasing transmittance. Further, the effect of transparency on brightness was greatest for test patch luminances near to those of their immediate backgrounds. The implications of these results for the understanding of configurational effects on brightness is discussed.

Introduction

Since the time of Helmholtz the influence of colored surrounds on the perceived color of test patches (i.e., 'induced' color) has been considered critical to our understanding of brightness, lightness and chromatic processing. In the Principles of Psychology, William James contrasted two views on induced color, the first championed by Helmholtz, and the latter by Hering.

12 Helmholtz argued that induced color was a "deception of judgment",

according to which observers judged the color of the test region as if it were illuminated by light with the same color composition as the surround. Thus, in the classic simultaneous brightness contrast display of Figure 1, in which a

Figure 1. Simultaneous Brightness Contrast. The gray patches are equiluminant.

gray patch on a bright background looks darker than it does on a black background, the visual system assumes that

the patch on the white surround is more highly illuminated. Given that the luminances of the two patches are nevertheless identical, the visual system infers on the basis of this assumption that the patch on the white background must be of lower reflectance, and that is how it is perceived. Hering, on the other hand, advanced a different view; he argued for a low-level physiological explanation for induced color in terms of lateral interactions at the retinal level. James summarized these two opposing views as follows:

Helmholtz maintains that the neural process and the corresponding sensation also remains

unchanged, but are differently interpreted; Hering, that the neural process and the sensation are themselves changed, and that the 'interpretation' is the direct conscious correlate of the altered retinal conditions. According to the one, contrast is psychological in its origin; according to the other, it is purely physiological.

William James (1890) The Principles of Psychology, 2, p. 19

The Helmholtz/Hering debate is in many ways cognate to recent controversy regarding the causes

of induced brightness phenomena, and the extent to which such phenomena reflect the influence of early-stage filtering operations as opposed to higher-order inferential processes is still unresolved.

6, 7, 11, 13,

20

Inferential mechanisms are essential to lightness perception, since the visual system must be able

to distinguish luminance discontinuities arising due to changes in reflectance from those that arise due to changes in illumination. The visual system accomplishes this under many circumstances with apparent

ease (as shown in Figure 2) since observers are frequently able to deduce that the lightness (i.e., reflectance) of a surface is uniform, even when partially covered by shadow, or occluded by a transparent overlay, where the regions may appear markedly different in brightness.

3, 4, 5, 16, 17, 19

Figure 2: Demonstration of perceived transparency.

While the computational necessity of inferential processes in lightness perception is uncontested,

2, 4, 8, 9, 10, 14, 17, 18 the role of perceptual inference in brightness coding is more problematic. Unlike

lightness, there is no a priori reason to expect perceptual inference to affect brightness. There are, nonetheless, several studies which suggest such an influence, and brightness changes attributed to inferential processes have been demonstrated for perceived stereo-depth,

18, 20 perceived pictorial depth or

shape,1, 14, 23

subject instruction in ambiguous displays,4 and perceived transparency.

1

The following experiment was performed to determine in a simplified, highly controlled stimulus, the extent

to which perceived transparency influences the brightness of a test patch, where perceived transparency was manipulated using both pictorial and stereoscopic depth cues.

Methods Subjects

Five subjects participated in the experiment, two of whom were completely naive regarding its

purpose. All subjects were well-practiced psychophysical observers and possessed normal spatial and stereo vision.

Stimuli

Stimuli were generated on a PC-compatible microcomputer (Vision Research Graphics, Inc.) and

presented on a 21” high-resolution display monitor. Frame refresh rate was 97 Hz (non-interlaced). Viewed from a distance of 60.7 cm the entire display subtended 32

o (w) by 24.2

o (h). Mean display luminance was

50 cd/m2. All images could possess 256 simultaneously presentable linearized intensity levels selected

from a palette of approximately 32,000. Stereo projection was achieved using pi-cell liquid crystal shutters synchronized to the monitor frame rate, such that alternate video frames were presented to the two eyes. In their open state, transmittance through the shutter glasses was 35%. Thus, viewed through the shutter glasses mean display luminance was 17.5cd/m

2.

The basic stimulus configuration is illustrated in Figure 3, which indicates the arrangement of the

test patch, the “inner” surround, the “outer” surround, and the matching patch common to all conditions. Test patches appeared on adjacent (inner) surrounds which were themselves made to appear either beneath transparent overlays, or as regions differing in reflectance from a wider (outer) surround.

Manipulated variables included the luminance of the test patch, and the luminances of the inner

surround as well as the upper and lower “wings” of the vertical (outer surround) rectangle. All other parameters were held constant. All luminances are expressed as percent maximum display luminance. Background luminance was fixed at 50% and the luminance of the right and left flanks of the outer surround was fixed at 80%. The square inner surround possessed five luminance values: 8%, 27%, 54%, 72%, and 80%. These were paired with five luminances of the upper and lower outer surround flanks (5%, 17.5%, 33%, 45% and 50%), such that in two configurations they were consistent with the interpretation of the stimulus as a vertical transparent rectangle overlying a horizontal rectangle which included the test patch.

Horizontal Rectangle

Outer Surround

(80% max luminance)

Vertical Rectangle

Outer Surround

(variable luminance)

Test Patch

Matching Patch

Inner Surround(variable luminance)

1o

6.7o

6.7o

1o

Background(50% max luminance)

13o

13o

12o

Figure 3: Stimulus arrangement.

The virtual transmittance values of the transparent vertical rectangle were 10%, 33%, 66%, 90%

and 100% (i.e. no transparent overlay). Test patch luminance ranged from 0% to 100%. Examples of the four stimulus configurations are illustrated in Figures 4-7. The first is the

transparency-with-stereo-depth-cues condition (set up here for crossed-fusion), in which the vertical transparent rectangle appears to float in front of the horizontal background rectangle, to which the test patch

Figure 4: Transparency-with-stereo-depth-cues.

is perceived to “belong”. In these examples the transmittance of the virtual transparency is 33%; test patch luminance is 32%.

The next condition is the transparency-without-stereo-depth-cues, in which pictorial cues alone suggest transparency, unreinforced by stereo cues.

Figure 5: Transparency-without-stereo-depth-cues.

The third condition is the no-transparency-without-stereo-depth-cues, in which the upper and lower “wings” of the vertical rectangle have been laterally displaced so as to invalidate pictorial cues to

transparency (i.e., eliminating the x-junctions). The final condition is no-transparency-with-stereo-depth-

cues where, in addition to the lateral offset of the upper and lower “wings”, stereo depth cues indicate that they lie behind the horizontal rectangle containing the test patch. In all four stimulus configurations the pattern of surround luminance is identical extending to a distance of 6.7

o from the center of the test patch,

making local contrast an unlikely explanation for the differential effects on brightness we observed.

Disparities were 0.3125o.

Procedure

We used the method of adjustment. Subjects were explicitly instructed to make brightness (not

lightness or brightness contrast) judgments, by matching the intensity of the test patch without regard to

other regions of the display. Test patch and surround luminances were presented in random order in conjunction with each transparency condition. Subjects completed either three or six adjustments per condition, from which the means and standard errors of the matches were computed.

Figure 6: No-transparency-without-stereo-depth-cues.

Figure 7: No-transparency-with-stereo-depth-cues.

Results

The pattern of data from all five subjects was very similar, and here we present means averaged

across observers. Figure 8 illustrates the pattern of mean match values (1 standard error). In each graph mean matching luminance is plotted as a function of test patch luminance for each of the four virtual transparency conditions (indicated by different symbols). The vertical dotted line on each panel indicates the luminance of the immediate test patch surround, dividing test patch decrements (to the left) from increments (to the right). The number in the upper left corner of each panel indicates the virtual transmittance of the simulated transparency. The fixed diagonal dashed lines (long dashes) represent the prediction for perfect luminance matching, while the variable-in-slope dashed lines (short-dashes) illustrate perfect ratio matching.

Consider first the 100% transparency condition. This is the condition with no transparent overlay or

no-transparency outer surrounds. Subject matches lay somewhere between perfect luminance matching (long-dashed line) and perfect ratio matching (short-dashed line). This is typical behavior for brightness matching in side-by-side displays.

22 Subjects also show the "crispening effect", in which brightness changes

most rapidly when test patch luminance is close to the luminance of its immediate surround. 21

Inspection of the results in the figures labeled 90%-10%, reveals that matching luminance

increases as the luminance of the inner surround decreases. This is indicated by the increase in slope of

the matching functions as the dotted vertical line (inner surround luminance) moves leftwards – that is, as the

100%

 0

 25

 50

 75

 100

10%

 0  25  50  75  100

33%

 0  25  50  75  100 0

 25

 50

 75

 100

66%

Test Patch Luminance (% maximum)

Ma

tch

ing

Lu

min

an

ce (

% m

axi

mu

m)

90%

 0

 25

 50

 75

 100

Transparency-with-Stereo-Depth-Cues

Transparency-without-Stereo-Depth-Cues

No-Transparency-without-Stereo-Depth-Cues

No-Transparency-with-Stereo-Depth-Cues

Luminance Matching

Ratio Matching

Background Luminance

Figure 8: Brightness matching data (average of five observers)

transmittance value decreases. It is clear, in addition, that while brightness matches in all cases still fall somewhere between luminance matching and ratio matching, decrement matches are in general much closer to the ratio matching prediction than are increment matches, which lie closer to the luminance matching prediction.

4 Most germane to this study, however, are the relative changes in test patch

brightness which occur as a function of the four stimulus configurations within each local background luminance condition. Note that brightness matches are generally highest, at all test patch luminances, in the transparency-with-stereo-depth cues condition (filled circles), and are generally lowest in the no-transparency-with-stereo depth cues condition (open symbols).

In Figure 9 we plot mean matching luminance (normalized to mean match value in the

transparency-with-stereo-depth-cues conditions) in the four stimulus configurations collapsed across subjects and test patch luminance. Virtual transmittance level is shown as a parameter in four separate panels. It is clear that the effect of transparency is greatest in the transparency-with-stereo-depth-cues condition and that this effect decreases with increasing transmittance. One-way repeated-measures ANOVAs performed on the mean luminance matches collapsed across subjects and test patch luminances indicated a significant (p<.05) effect of stimulus configuration at all transmittance levels. Post-hoc comparisons of the mean luminance matches at all transmittance levels revealed that mean matching luminance in the transparency-with-stereo-depth-cues condition (configuration condition 1) was significantly higher than in all other conditions. In addition, for the 10% transmittance condition, mean luminance matches in the transparency-without-stereo-depth-cues condition were significantly greater than in the no-transparency-with-stereo-depth-cues condition. No other pairwise comparisons were significant.

In order to assess the magnitude of the effect on brightness which the transparency-with-stereo-

depth-cues condition exerted relative to that produced by simple brightness induction, the change in mean brightness matches in the transparency-with-stereo-depth-cues condition was assessed relative to the average brightness matches in the other three conditions. Percent brightness change is plotted as a function of the ratio of test patch luminance to inner surround luminance in Figure 10. Virtual transparency transmittance appears as a parameter.

10%

 80

 90

 100

Transparency-witn-Stereo-Depth-Cues

Transparency-without-Stereo-Depth-Cues

No-Transparency-without-Stereo-Depth-Cues

No-Transparency-with-Stereo-Depth-Cues

33%

66%

Configuration Condition

1 2 3 4Me

an

Ma

tch

ing

Lu

min

an

ce

(%

Tra

nsp

are

ncy-w

ith

-S

tere

o-D

ep

th-C

ue

s c

on

fig

ura

tio

n)

 80

 90

 100 90%

1 2 3 4

Figure 9: Brightness matches collapsed across test patch luminance.

Two features of these data are notable. First, across all transmittance values the largest brightness enhancements occur for test patches whose luminances are very near to that of the inner background luminance (that is, for ratios of test patch to inner background luminance near 1.0). Second, brightness enhancement increases with decreasing transmittance where, at maximum in the 10% transmittance condition, test patch brightness is enhanced by transparency by slightly more that 50%.

Discussion The purpose of this study was to determine whether, and to what degree, perceived transparency in

the absence of confounding local luminance cues affected the judgment of brightness. Transparency-with-stereo-depth affected brightness such that subjects perceived the test patch to be brighter than it would otherwise have appeared. This result is consistent with Adelson’s

1 hypothesis that the

brightness of the test patch is in part determined by the brightness it would possess were the transparent overlay to be discounted. A possible explanation as to why this effect manifested itself predominantly in the stereo-depth conditions is that only under these circumstances is there an unambiguous and compelling impression of transparency. The ordinal pattern of mean brightness matches of the previous slide supports

this interpretation, since brightness matches are consistently highest in conditions containing both stereo and pictorial cues to transparency, are next highest in conditions containing only pictorial depth cues, and

are lowest in conditions where both stereo and pictorial depth cues falsify the perceptual hypothesis of transparency. According to this view one might also expect a brightness effect, even in the absence of stereo-depth cues, by using a more complex background such as a Mondrian (or the Argyle background). The additional information afforded by a complex background, one containing more numerous x-junctions and luminance ratios consistent with physical transparency, might render transparency a more parsimonious (and hence more compelling) perceptual interpretation, thus leading to a more robust brightness effect.

Transparency-with-Stereo-Depth-Cues versus Other Three Configurations (mean)

Test Patch / Background Luminance

0.1 1 10

Me

an

Ma

tch

ing

Lu

min

an

ce

(%

ch

an

ge

)

-10

0

10

20

30

40

50

6010% transmittance

33% transmittance

66% transmittance

90% transmittance

Figure 10: Percent change in normalized brightness matches as a function of test patch/background luminance ratio.

Also noteworthy is the fact that the effect which transparency exerts on brightness increased with

decreasing transmittance, and that it was maximal for test field to background luminance ratios near unity (100%). This ratio corresponds to the region where brightness changes most rapidly with changes in test field luminance, resulting in the "crispening effect."

21, 22 It is also the region where luminance discrimination

thresholds are smallest. 21, 22

Thus, since the effect of transparency on brightness is relatively modest, at least as compared to that of local luminance, its expression might be expected to be maximal where the sensitivity of the brightness system is highest. Within this context it is worth noting that in the Argyle stimulus the luminance ratio of the diamonds to the mean background luminance is 107% -- very near the optimal ratio. Thus, the luminances of the diamonds lie within that narrow regime identified in our experiment over which perceived transparency most strongly influences brightness. One prediction from our results is that if the underlying reflectance map is altered such that the luminance ratios of the diamonds to the background, as seen through the virtual transparency, lie outside that identified regime, then the effects of transparency on brightness should be weakened.

This prediction is tested qualitatively in the demonstration of Figure 11. The upper row (left)

illustrates the reflectance map of the Argyle stimulus. The middle panel illustrates Adelson’s Argyle stimulus and the right panel illustrates Adelson’s control stimulus, in which the impression of transparency in the experimental stimulus is defeated by introducing gray gaps between the columns of diamonds. Note that it is the difference in brightness between adjacent columns of diamonds in the experimental versus control conditions which indexes the effect of perceived transparency. The lower row (left) illustrates the reflectance map of a modified Argyle stimulus in which the columns of diamonds now possess equal reflectance. The middle panel is the modified reflectance map as viewed behind the same transparent

squarewave grating as the Argyle stimulus, above. The right panel is the modified Argyle control condition in which transparency is again defeated by separating the columns of diamonds with gray gaps. In the modified Argyle stimulus the luminance ratios of the dark and bright diamonds to the mean background luminance are 77% and 110%, respectively. This departure from the optimal test/background luminance ratio of approximately 100% should and does weaken the effect of transparency on brightness, which inspection of Figure 11 confirms. Thus, relatively low level factors such as test/background luminance ratio, in addition to higher inferential processes, mediate the effect of transparency on brightness.

Original Argyle

Modified Argyle

Reflectance Map Experimental Stimulus Control Stimulus

Reflectance Map Experimental Stimulus Control Stimulus

Figure 11: Transparency effects on brightness in Adelson’s original and a modified Argyle stimulus.

In conclusion, the results of this study reinforce the idea that, in addition to early filtering

mechanisms, brightness also depends to some extent on the operation of higher-level inferential processes involved in segmenting scenes into a representation of reflective surfaces and illuminations, but only under a restricted range of stimulus conditions. In particular, our results point to a prominent role for stereo-depth supported transparency in this process.

15

References

1. Adelson, E. H. (1993) Perceptual organization and the judgment of brightness. Science, 262, 2042-2044. 2. Agostini, T. and Proffitt, D. R. (1993) Perceptual organization evokes simultaneous lightness contrast. Perception, 22,

263-272. 3. Arend, L. E. and Goldstein, R. (1990) Lightness and brightness over illumination gradients. Journal of the Optical Society

of America, A, 7, 1929-1936. 4. Arend, L. R. and Spehar, B. (1993) Lightness, brightness and brightness contrast: 1. illuminance variation. Perception &

Psychophysics, 54, 446-456. 5. Arend, L. R. and Spehar, B. (1993) Lightness, brightness and brightness contrast: 2. reflectance variation. Perception &

Psychophysics, 54, 457-468.

6. Blakeslee, B. and McCourt, M. E. (1996) Does a single mechanism underlie simultaneous brightness contrast and grating induction? Invest. Ophthalmol. Vis. Sci. (Suppl.), 37, S1066.

7. Blakeslee, B. and McCourt, M. E. (1996) Similar mechanisms underlie simultaneous brightness contrast and grating induction. Vision Research (submitted for publication).

8. Buckley, D. Frisby, J. P. and Freeman, J. (1994) Lightness perception can be affected by surface curvature from stereopsis. Perception, 23, 869-881.

9. Cataliotti, J. and Gilchrist, A. (1995) Local and global processes in surface lightness perception. Perception and Psychophysics, 57, 125-135.

10. Gilchrist, A. L. (1977) Perceived lightness depends on perceived spatial arrangement. Science, 195, 185-187. 11. Gilchrist, A.L. (1996) Lightness contrast and lightness constancy: what is the relationship? Chapter in forthcoming book

by A. Gilchrist, (Ed.) (title unavailable at present). 12. James, W. (1890) The Principles of Psychology. Dover Publications Inc., New York. 13. Kingdom, F.A.A., McCourt, M.E. and Blakeslee, B. (1996) In defense of "lateral inhibition" as the underlying cause of

induced brightness phenomena. A reply to Spehar, Gilchrist and Arend. Vision Research, (in press). 14. Knill, D. C. and Kersten, D. (1991) Apparent surface curvature affects lightness perception. Nature, 351, 228-230. 15. Knill, D. C., Kersten, D. and Mamassian, P. (1996) Implications of a Bayesian formulation of visual information processing

for psychophysics. In: Knill, D.C. & Richards, W. (Eds.) Perception as Bayesian Inference. Cambridge University Press, Cambridge, MA.

16. Metelli, F. (1974) The perception of transparency. Scientific American, 230, 91-98. 17. Schirillo J., Reeves A. and Arend L. (1990) Perceived lightness, but not brightness, of achromatic surfaces depends on

perceived depth information. Perception and Psychophysics, 48, 82-90. 18. Schirillo, J. A. and Shevell, S. K. (1993) Lightness and brightness judgements of coplanar retinally noncontiguous

surfaces. Journal of the Optical Society of America, A., 10, 2442-2452. 19. Sewall, L. and Wooten, B. R. (1991) Stimulus determinants of achromatic constancy. Journal of the Optical Society of

America, A., 8, 1794-1809. 20. Spehar, B., Gilchrist, A. and Arend, L. (1995) The critical role of relative luminance relations in White's effect and grating

induction. Vision Research, 35, 2603-2614. 21. Whittle, P. (1992) Brightness, discriminability and the "crispening effect". Vision Research, 32, 1493-1507. 22. Whittle, P. (1994) Contrast brightness and ordinary seeing. In: Gilchrist (Ed.), Lightness, Brightness, and Transparency,

(pp. 111-157). Hillsdale: Erlbaum. 23. Wishart, K. A., Frisby, J. P. and Buckley, D. (1996) The role of 3-D surface slope in a lightness/brightness effect. Vision

Research, (in press).

Acknowledgments

MEM was supported by grants from the National Eye Institute (EY10133-01) and the Air Force Office of Scientific Research F49620-94-1-0445). BB was supported by grants from the National Science Foundation (IBN-9514201 and IBN-9306776). FAAK was supported by a grant from the Medical Research Council of Canada (MT11554).

X =

Background Overlay Multiplicative

Transparency

Figure 6: Missing-fundamental Argyle Illusion.

The Argyle illusion,

1 shown in Figure 3, is regarded as one of the most compelling demonstrations

of the effect which transparency may have on brightness perception. Here, the three columns of diamonds in the left-hand panel are equiluminant, yet appear unequally bright. In an argument similar to that advanced vy Helmholtz, Adelson (1993) proposed that, through a process of discounting the illuminant, the diamonds beneath the dark bars are inferred to possess a higher reflectance than those seen beneath the bright bars of the squarewave, and that it is the differential lightness of the diamonds which influences the brightness judgment, such that brightness assimilates to lightness. Adelson’s control stimulus is shown in the right-hand panel of Figure 3. In this stimulus gray gaps have been inserted to either side of the dark bars of the squarewave transparency which disrupts the impression of transparency. Thus, the effect of transparency on brightness is indexed by the difference in the brightness between adjacent columns of diamonds in the experimental versus control conditions of the illusion. The manipulation used to disrupt transparency in this

Experimental Stimulus Control Stimulus

Figure 3: The Argyle illusion.

case also does also change slightly the pattern of local luminance in the vicinity of the diamonds, where these luminance cues would work to reduce the estimated magnitude of the transparency effect.