Measuring Constancy of Contrast Targets in Different Luminances in Complex Scenes

Objects in complex images appear almost constant. This is true with changes of the overall level of illumination, and somewhat true with changes of the surround around a test area. The small departures from perfect constancy, however, provide important evidence on the underlying mechanisms of constancy. First, this paper measures the departures from constancy with changes in overall luminance. Second, it measures the effects of contrast using white, gray, and black surrounds. Third, it compares the results from flat-2D transparent displays with those using 3D shapes. With 3-D paper targets, illuminated in direct light and shadow, we find the same small decrease in matching value with large decreases in illumination level (low-slope behavior) found in flat-2D transparent displays. Within the direct light and in the shade parts of the targets, the matches showed the same high-slope contrast behavior. Here, the arrangement of reflectances, illumination and depth did not affect the appearance matches made by observers. In both flat-2D transparent and complex 3-D data, observers’ matches fit the simple two-step physical description. The local maxima are dependent on luminance, and other, darker areas, are dependent on spatial contrast. Introduction One of the first reported psychophysical experiments was the classification of stellar magnitude by Hipparchus of Nicea in the 2 century B.C. Although the original manuscripts have been lost, the results were documented by Ptolomy. After many centuries, stellar magnitude is in common use today. It has been modified to be a photometric measurement starting with Pogson in 1856. The stellar magnitude changes by 100:1 when the measured luminance changes by 100,000:1. In other words, stellar magnitudes have a slope of 0.4 in a plot of log luminance vs. stellar magnitude. Flat-2D transparent Displays In our flat-2D-transparent study, observers matched the appearance of eight white, gray, and black patches using different illuminances. The experiment asked seven observers to match uniform luminance patches to a standard display. This experiment included 3,150 observations. Both test and standard displays were transparent photographic films viewed on two high-luminance Aristo lightboxes. The targets subtended 25 by 30 degrees. The two light boxes were mounted at right angles so that the observers turned their heads back and forth to make matches. The left eye was used for observing the contrast targets, and the right eye for observing the standard. The standard was a series of 9 patches each with a different luminance transmission. The nine patches were surrounded by white (1000 ft-L). The patches were selected to have equal differences in appearance between white [9.0] and black [1.0]. In the experiments described below, the observers were asked to report on the mixture of colors on a palette that would match what they saw. Observers were asked to interpolate between the 9 reference values seen in the standard display. The dynamic range of the test targets covered 2.3 log units; the dynamic range of the illumination covered 3.3 log units. The resulting dynamic range of the experiment covered 5.6 log units. One advantage of this apparatus is the luminance uniformity across the display and within each test patch. The luminances reported in this paper were measured with a Gamma Scientific telephotometer. Another advantage of using photographic transparencies is that we can be certain that the contrast displays are exactly constant at each illumination level. The ranges of the transparent contrast targets are larger because they are not limited by paper surface reflectances. The constancy of image content is far superior to what is possible in simulations on monitors and LCD displays. This study investigates the role of luminance and contrast in complex images. It uses eight different transmission patches (2.7° by 5.5°) test patches. in a white, a gray and a black surrounds. These patches are viewed at 5 different luminances and matched to a constant standard. The goal of the study is to understand the limits of constancy with overall changes in luminance. The experimental data will be compared to various hypotheses such as “discounting the illuminant”, simultaneous contrast, and the ancient low-slope change of appearance with luminance found in stellar magnitude. A further goal is to expand the range of luminances matched in related studies. Figure 1 plots the observer matches for 8 test areas in a white surround. The lines connect the average match for 7 observers for the same area in 5 different illuminations We used a least squares regression linear fit to the data from each line in a white surround to calculate the linear slopes of the eight patches. The average slope for a white surround is 0.500 ± 0.108. When analyzing the data for the same luminance test areas in a gray surround the average linear fit for the eight patches in gray surround was 0.503 ± 0.108. The average linear fit for the black surround was 0.612 ± 0.066. These average slopes are plotted in Figure 2. The average linear fit for all lines in all surrounds is 0.539 ± 0.103. Figure 1 plots the match for 8 test areas in a white surround. Here, the different lines plot the matches for each patch in the contrast display. The eight lines for each target area are parallel lines with a very low slope Figure 2 plots the average slope for each of 8 patches in 5 illuminants. Here we see the same behavior for white, gray, and black surrounds. The fact that matches in different surrounds exhibited the same low-slope rate of change with luminance give an important insight into the effect of overall illumination. Decreasing illumination departs from perfect constancy, but at a slow rate. That rate is the same regardless of the effects of spatial contrast. That rate is the same for all test patches in all surrounds. Changes in overall luminance adjust all parts of the image so that the maxima in the scene fall on the 0.54 slope “Hipparchus line”. Figure 3 replots the same data used in Figure 1. Here the lines connect the average match for 8 test areas in each illumination. These are the contrast lines. The contrast plots of observer data show 5 parallel lines, one for each illumination level. We fit the data from each line in Figure 3 with linear least square regression to calculate the contrast slopes. The values were: 4.70, 4.75, 4.46, 4.34, 4.82. The average slope for a white surround is 4.615 ± 0.206. Figure 3 plots the match for 8 test areas with a white surround. The lines show the matches for the 5 levels of lightbox luminance. The five contrast