Printer Friendly

Visual saliency of urban objects at night: impact of the density of background light patterns.


Visual salience is the perceptual quality which makes some items in a scene stand out and immediately grab attention [Itti and others 2002; Itti 2005]. Traditionally, in urban lighting, increasing luminance has been one of the main tools used to improve the saliency of objects. However, insufficient guidance and planning control can cause competition for illuminated urban objects to be more visible than their neighbors, creating a light war and visual anarchy in the nightscape. Existing guidelines (for example, the ILE Outdoor Lighting Guide [ILE 2005]) tend to recommend a luminance ratio between an object and its background according to the degree of conspicuity (saliency) required, with a higher luminance contrast being recommended for a higher degree of conspicuity.

These recommendations are generally interpreted to be based on the conspicuity of an object in front of a uniform background. In real urban situations objects tend to have complex backgrounds.

The saliency of an object relies on properties of the context in which the target is embedded in addition to properties of the object such as local feature contrast (for example, color contrast and brightness contrast.) [Adrian and Eberbach 1968, Aks and Enns 1996; Bloomfield 1972; Eckstein and others 1997; Paulmier and others 2001, Nothdurft 1992]. A number of studies have shown the importance of context factors in object detection and recognition [Palmer 1975; Biederman and others 1982; Chun 2005; Wolfe and others 2002; Gibson and Jiang 2001]. For example, Biederman and others [1982] show that when a cylindrical object is placed in a kitchen it is more likely to be recognized as a breadbox rather than a drum. Palmer shows that the positions of objects within scenes are also important, making it difficult, for example, to detect sofas floating in the sky or fire hydrants sitting on top of mailboxes [Palmer 1975].

In classifying light, there are several descriptions two dimensional patterns of light on surfaces [for example, Cuttle 1971; Nassar 1998; Gordon 2003]. Cuttle's highlight patterns and Gordon's focal glow refer to the two dimensional effect of lighting, which in this study is regarded as patterns of light. In the current study background distracting objects are elements which present an abrupt gradient from high to low brightness against the background, such as windows to an illuminated interior space and light fittings. Uniformly illuminated surfaces were not assumed to be light patterns.


Previous work suggests that increasing the number ofobjects in the background reduces the conspicuity of a target [for example, Nothdurft 2000; Bloomfield 1972; Turatto and Galfano 2000; Jenkins and Cole 1982; Eriksen and Eriksen 1974; Smith 1998; Wolfe and others 2002]. For example, using abstract objects Eriksen and Eriksen [1974] and Smith [1998] demonstrated that when the target object differs in shape from other objects, increasing the density of nontarget objects (that is, increasing the number of nontarget items within a constant search area) reduces the conspicuity of the target. Similar results occur when stimuli of varying color, shape, size, and contrast are used as nontarget stimuli [Bloomfield 1972].

A single salient target is found to attract attention independent of the number of other, nonsalient elements in the background [Nothdurft 1993; Treisman and Gelade 1980]. In more complex cases, the number of nontargets in the display has a large effect on performance, though in fact, this figure varies widely and continuously from one task to another [Desimone and Duncan 1995]. For example, increasing the density of nontarget objects increases the salience of objects which are different from their background by orientation and motion contrast, but not when they are different by luminance and color contrast [Nothdurft 2000].

Jenkins and Cole show that increasing the number of background elements has a significant effect on target saliency when the basis of discrimination is luminance difference [Jenkins and Cole 1982]. In their study they show that increasing the density of background disks from 5 to 10 percent of the background area has a significant decrease on the saliency of the target disk (increasing the density included increasing in the area occupied by discs and also the number of discs in the background). There was, however, no further decrease of performance when background density was increased to 15 percent. The effect suggests diminishing returns. They also show that increasing the density of background disks increased the luminance contrast necessary for the target disk to be seen when the target differed from the background elements in luminance.

The literature suggests that the density of background objects can affect the visual saliency of target objects, and that this effect can vary depending on the situation. These studies tend to consider targets of size or luminance below or around visibility threshold levels.

In this study, further experimental work was carried out to extend evidence of background density and saliency:

* the saliency of targets which are in supra threshold visibility levels

* real objects which are already prominent features in their context, that is, urban monuments and landmarks.

* background objects are patterns of light in this study.

A further article investigates the effect of proximity of background objects to the target.



A pilot study was carried out to validate proposed levels of the density of background patterns of light, that is, whether the differences between levels of background density were apparent to observers. Saliency was examined using a forced choice task comparing the saliency of objects in pairs of images.

Figure 1 shows the images used in the pilot study. These are black and white photographs of two urban scenes at night time, digitally manipulated using Adobe Photoshop to create four levels of background density. Images have been used as environmental presentation medium over wide range of research principals and assume to be an efficient surrogate for real environment settings [Coeterier 1983; Shuttleworth 1980].

This study investigated the effect of background density on saliency and thus required a comparative measure of the density of background light patterns. For these images of fixed area, density can be determined by the number of background objects or the area occupied by these objects; the former says nothing about the size of the background objects and the latter says nothing about the number of discrete items. Thus both methods were employed to provide concurrent validation.

In the box counting method a grid of squares was superimposed onto the test images and the number of squares containing light patterns was used to define the density of light patterns, similar to the method used by others [Bovill 1996; Carrera 1997]. The light patterns were distributed as evenly as possible across the image to avoid spatial concentration of the light patterns in any one area, although this was judged only by eye. Where a light pattern, such as a wall or a large window, was of an area larger than a single square on the grid it was treated as a lit surface and not categorized as a light pattern. If a single pattern extended across two or more grid squares it was counted only once. The box counting method is similar to the approach used by Wager and Heisler who placed a matrix of dots on an acetate sheet placed over the photograph of a tree crown to quantify tree crown density and compared the proportion of dots covering tree crown to those covering the sky [Wager and Heisler 1986]. This area occupied by background light patterns was determined using a count of the number of pixels and this was done using Adobe Photoshop.


Four levels of background density (BGD) were defined: no background patterns of light, and low, medium and high densities of background patterns. The medium density image had 50 percent more background light patterns than did the low level (according to the box counting method) and the high level 50 percent more than the medium level. By increasing the pixels ratio and box number ratio in a same rate aimed to have the same percentage ratio of pixels and box numbers. For example;

20 percent / 50 percent [number of boxes] = 2 percent / 5 percent [number of pixels].

Each box comprises approximately 24 X 24 pixels in Scene 1 and 23 X 23 pixels in Scene 2. From the original photographs light patterns were digitally deleted or some extra light fixtures or patterns of light on urban envelops were added to create the different levels of background density.

Table 1 shows the background densities of these images as characterized by the box counting and pixel counting methods. For the pilot test there were no intentional target objects in these images.


A rating task was used to determine whether subjects could discriminate between the four different levels of background density. The eight images were presented individually and in a random order. Test participants were instructed to rate the density of light patterns in the picture using a six point scale from 1 (lowest density) to 6 (highest density). Examples of high and low density of background patterns from both images were shown to the test participant prior to the test to anchor the stimulus and response ranges. No time restriction was imposed for this task and the typical time for completion was less than five minutes. Nine volunteer test participants were used and these were students of School of Architecture at The University of Sheffield.


The results of the pilot study are shown in Table 2. The ratings tend to agree with the intended level of background density, with higher ratings being given to the images of higher density. The Friedman test was applied to the results of each scene separately, and this suggests that there are significant differences in ratings for the four levels of background density (p < 0.01). The Wilcoxon test was applied to compare the ratings of adjacent density pairs and this also suggests significant differences in ratings (p < 0.01). The results also suggest the ratings given by different test participants for a given image were of reasonable agreement and this is confirmed using Kendall's W test (p < 0.01). These results suggest that the proposed classification of background density creates images for which the intended differences in background density are apparent to test participants. The four proposed levels of background density were therefore used in the main study investigating background density and saliency.



A forced choice pair comparison test was carried out to rank the visual saliency of urban objects observed with different levels of luminance contrast and levels of the density of background patterns of light.


The test used the two black and white images of urban scenes at night time as were used in the pilot study; for these trials the target object was present. Scene 1 subtended a visual field of 24[degrees] width and 18[degrees] height at the observers eye; Scene 2 was 26[degrees] wide and 20[degrees] high. In both images an urban sculpture was used as the target as shown in Fig. 2. The target objects were not identical but they did present the same width (1[degrees]) and height (3[degrees]), a size sufficiently large to gain attention at first glance. The images were presented using a pair of identical computer monitors (Viglen VD 695, 15 inch screen, resolution 1280 X 1024 / 60 Hz, Dot Pitch / Pixel Pitch 0.24 mm). Microsoft PowerPoint was used to present the images.

Saliency was examined using images presenting four levels of background density and four levels of target-background luminance contrast: the two urban scenes were digitally manipulated to present all 16 possible combinations of luminance contrast and background density. Sample images are shown in Figs. 3 and 4.

Luminance contrasts of 0, 3, 5, 10 were used, these being suggested to give effects that are not noticeable, just noticeable, low drama, and high drama respectively [ILE 2005]. Luminances were determined using pixel brightness values from Photoshop [Moore and others 2000; Kimura and Noguchi 2002; Hagiwara and others 2004]. Luminance contrasts were determined using the standard expression [C.sub.L] = ([L.sub.T] - [L.sub.B])/[L.sub.B]. where [L.sub.T] is the average pixel brightness of the target and [L.sub.B] is average pixel brightness of the whole scene other than the target. The mean pixel brightness of target background in different levels of background density is shown on Table 3. While it was attempted to maintain similar background luminances for all variations of background density it can be seen that the background luminances do vary slightly. Table 4 shows the actual luminance of the target and its immediate background as well as the darkest area in the Scenes for each level of luminance contrast.

The tests were carried out in a laboratory in which the artificial lighting was not switched on and daylight excluded using blinds. A 40W GLS lamp was used to provide a low level of background lighting, similar to night-time conditions, and this was located to avoid glare on the subjects' eyes or reflections on the display screens. This background lighting produced an average illuminance of 5.6 lx on the front of the table upon which the PC displays screens were placed.

Thirty volunteer subjects were used. These were mostly students of the School of Architecture, University of Sheffield, and 18 were male. The age range was between 25 and 45 years old and all reported normal or corrected vision. Half of the sample carried out tests using only Scene 1, the other half seeing only Scene 2.

A forced choice task was employed, in which two test images were presented in juxtaposition and test participants reported in which image the target appeared more salient. Both spatial and temporal image juxtaposition was used in separate trials and each test participant used both methods.



For spatial two-interval forced choice, two test images were presented simultaneously on side-by-side monitors and test participants reported whether the left-hand or right-hand monitor presented the image with the more salient target. The predicted number of left and right responses was balanced to counter a stimulus frequency bias.

Twenty minutes adaptation time was allowed before trials, during which time the test instructions were delivered. A test commenced with between five and ten practice trials, this continuing until the subject felt confident with the procedure.

Pairs of images were shown for maximum of 10 seconds following black screen intervals between pairs of images.

For temporal two-interval forced choice, two test images were presented one after another on a single monitor and test participants reported whether the first or second image presented the more salient target.

Each test image was presented for three seconds; before each test image a neutral image [Poulton 1989] (that is, test scene but without a target object) was presented for one second. This procedure is illustrated in Fig. 5.

The 16 different images of the each scene were compared in all possible pairs, thus requiring 120 comparisons to be made.

To reduce sequential contraction bias, The presentation order was randomized and the expected results for left and right monitors in spatial method and first and second sequence in temporal method were counterbalanced [Poulton 1989]. However there could be still bias caused by positional preferences towards either left or right monitors in the spatial method and also preferences to 1st or 2nd intervals in the temporal method [Jakel and Wichmann 2006]. To examine the effect of any bias, null condition trials were included. In these trials the same image was presented in both monitors in the spatial method and in both first and second presentation of the pair in the temporal method. The results show no bias towards any of the monitors in spatial method and no bias towards any of the sequences in temporal method [Davoudian 2010].




In the trials, a score of one was given to the image reported to be more salient and a score of 0 given to the second image. A summation of these results for each of the 16 variations of a test image provides a saliency value, with a high saliency value indicating an image in which the target appears to be more salient than images with lower saliency values.

The Mann-Whitney test does not suggest any significant differences between the results obtained from two scenes and also from the two methods, thus further analyses were carried out for the combined results. Data for both images combined and the results are shown in Fig. 6. It can be seen that, as expected, a target tends to become more salient as the target-background luminance contrast increases, and the target becomes less salient as the density of background light patterns increases. Within each level of luminance contrast, the Friedman test suggests that background density has a statistically significant effect on saliency (p < 0.01).


Saliency values in different levels of background density are illustrated in Fig. 7 and Table 5, and this suggests a trend for saliency to decrease as the density of background light patterns increases. Friedman's test suggests significant differences of saliency between the four levels of background density, p < 0.05, as summarized in Table 6.


Wilcoxon signed rank test was used to examine whether the differences between the saliency value of two immediate BGD level is significant. The results are shown on Table 7.

The results show a significant differences between the saliency value of the immediate levels of BGD, p < 0.05, apart from High and Med level of BGD when the luminance contrast is 10. When differences were statistically significant, an approximate effect size (r) was calculated to standardize the magnitude of the effect observed. The Z score was converted into estimated effect size using the following equation in which N is number of observations [Rosenthal 1991];

r = Z/[square root of N]


Table 8 shows the effect size of the test. It can be seen that the value of effect size increases by increasing the level of BGD between No-Low and Low-Med. However, this value drops considerably between Med and High level of background density. This trend is not the same in [C.sub.L] = 10 which the highest effect size is between No-Low level of BGD and there is a slight decrease in effect size. Interestingly, by increasing the luminance contrast the magnitude of effect also increases, however, when BGD increases from Med to High it shows a reverse effect in some cases. This result implies that by increasing the background density of light patterns, the impact of density on saliency value will not stay constant and after a certain level the effect of BGD reaches maximum level, and it may result to no more effects.


Spearman's test was used to examine whether these changes in saliency value imply a correlation between density of background lighting and saliency value. The regressive trends of votes by increasing the background density could be seen in all the levels of luminance contrast apart from [C.sub.L] = 0. Data from [C.sub.L] = 0 shows no significant differences between different levels of Background Density of light patterns (BGD) and the results are not significantly different in BGD levels, therefore these data were excluded from the correlation analysis.

The results of correlation analysis are presented in Table 9. The results show that saliency value has a significant negative correlation with the background density of light patterns, p < 0.05. The correlation between luminance contrast and saliency value was also tested. As expected there is a significant relationship between saliency value and the level of luminance contrast, p < 0.05. This suggests that BGD accounts for 26.1 percent of variance in saliency value, while luminance contrast accounts for 35.8 percent of variances in saliency value.

Interaction between the variables (luminance contrast and background density) could affect the result such that the derived effect size might not be representative of the real impact of BGD. Thus, partial correlation test was adopted. Using partial correlation allows the relationship between two variables to be measured while controlling for the effect of that one or more additional variable has on one of those variables. Here the relationship between saliency value and BGD was tested while the impact of luminance contrast was controlled. Partial correlation suggests that BGD accounts for 39.8 percent of variance in saliency value, p < 0.05. See Table 9.

Here the correlation test is carried out on the excluded data in no contrast condition, [C.sub.L] = 0. The aim is to examine whether there is any correlation between saliency value and BGD in this condition. Surprisingly, against the negative correlation in other conditions the results show a significant positive correlation between variables, p < 0.05. A possible explanation of this result could be found in local contrast effect by increasing the number of light patterns around a nonilluminated target. The local feature contrast increases the visual attention on the target regardless of luminance contrast of the target/ background.


These data suggest that the density of light patterns in the surroundings of the urban object affects the visual saliency of urban objects.

This study supports previous studies on the impact of density of background objects [for example, Nothdurft 2000; Bloomfield 1972; Turatto and Galfano 2000; Nothdurft 1993; Jenkins and Cole 1982] and shows a reverse correlation between increasing the density of light patterns and saliency level of the illuminated urban object. However the current study involved supra-threshold visibility levels allowed by higher luminance contrast; the luminance contrast level of the urban object did not affect the results which are consistent with those of Jenkins and Coles [1982] who used luminance contrast levels either below conspicuity thresholds or just beyond the thresholds. It is also found that the impact of density of light patterns is not constant between the different density levels; the impact decreases for higher densities of light patterns but after a certain level further increases have no effect.

Results of this study simultaneously confirm and opposes the results of studies on pedestrian conspicuity at night. Moberly & Langham [2002] show that visual clutter could reduce the visibility of pedestrians at night and Tyrrell and others [2009] show that visual clutter do not reduce the visibility of pedestrians at night. The limitation in both studies is that the density and distribution of clutter in regard to target pedestrians have not been defined clearly and this factor can significantly affect the results. For this reason it was tried to apply a systematic classification on background density of light patterns in this study to make the judgment for the future studies much easier.

The identification of the impact of background lighting design factors on the appearance of urban objects in this study makes it possible to suggest that the significant lighting factors go beyond those normally presented elements in lighting literature such as luminance and color contrast and uniformity or nonuniformity of the background lighting. Guidelines such as the ILE Outdoor Lighting Guide [2005] could consider the result of this study in future versions. The lighting context could be more fully addressed by considering the number and size of light sources in the outdoor area to be lighted.

These findings should also inform and motivate development of future models of visual recognition in the road environment. Such models should address the important effects of environmental context in addition to the photometric variables (such as luminance and contrast) that are the only factors considered in traditional models of visibility level.

One might want to examine salience of other types of targets--like pedestrians, and perhaps the effects of target motion or color. Also, while planning to move on to new lighting systems (as will occur with technology developments), new lighting systems should not result in additional clutter in the lighting context.

It should be noted, however, that psychophysical research methods are sometimes criticized for isolating particular sensations and consequently lacking the possibility to generalize the results to wider real-world conditions [Jay 2002]. It is acknowledged that the results of this research relate only to an interior experimental set up that may not be a fair representation of the onsite experience of subjects.


On the basis of the results, the existing assumption about luminance contrast level of object/background is conservative in its estimate for saliency level of urban objects in complex lighting background. Including properties of lighting context such as complexity factors-including density of background light patterns--in the assumption of saliency level should make the prediction of visual appearance of urban objects more accurate. This study demonstrated that background lighting complexity is influential in the saliency of urban objects at night. Guidelines based solely on luminance contrast are not sufficient.

doi: 10.1582/LEUKOS.2011.08.02.004


Adrian W, Eberbach, K. 1969. On the relationship between the visual threshold and the size of the surrounding field. Lighting Res Technol. 1(4):251-254.

Aks DJ, Enns JT. 1996. Visual search for size is influenced by a background texture gradient. J Exp Psychol Hum Percept Perform. 22(6):1467-1481.

Biederman I, Mezzanotte RJ, Rabinowitz JC. 1982. Scene perception: detecting and judging objects undergoing relational violations. Cogn Psychol. 14(2):143-177.

Bloomfield JR. 1972. Visual search in complex fields: size differences between target disc and surrounding discs. Hum Factors. 14(2):139-148.

Bovill C. 1996. Fractal Geometry in Architecture and Design. York (PA): Birkhauser Boston. 195 p.

Carrera F. 1997. Campo Santa Maria Formosa, Venice, Italy: A case study of the application of visual, dynamic and scale-invariant analyses for the description, interpretation and evaluation of city form. [term paper]. [Boston (MA)]: Massachusetts Institute of Technology.

Chun MM. 2005. Contextual guidance of visual attention. In: Itti L, Rees G, Tsotsos JK. eds. Neurobiology of Attention. London (UK): Elsevier Academic Press. 744 p.

Coeterier JF. 1983. A photo validity test. J Exp Psychol. 3(4):315-323.

Cuttle C. 1971. Lighting patterns and the flow of light. Lighting Res Technol. 3(3):171-186.

Davoudian N. 2010. The Impact of background lighting complexity on the visual saliency of urban objects at night. [dissertation]. [Sheffield (UK)]: University of Sheffield.

Desimone R, Duncan J. 1995. Neural mechanisms of selective visual attention. Ann Rev Neurosci. 18:193-222.

Eckstein MP, Ahumada AJ, Watson AB. 1997. Visual signal detection in structured backgrounds. II. Effects of contrast gain control, background variations and white noise. J Opt Soc Am A Opt Image Sci Vis. 14(9):2406-2419.

Eriksen BA, Eriksen CW. 1974. Effect of noise letters upon the identification of a target letter in a nonsearch task. Percept Psychophys. 16:143-149.

Gibson BS, Jiang Y. 2001. Visual marking and the perception of salience in visual search. Percept Psychophys. 63(1):59-73.

Gordon G. 2003. Interior Lighting for Designers, 4th ed. Hoboken (NY): John Wiely & Sons. 312 p.

Hagiwara T, Kizaka K, Fujita S. 2004. Development of visibility assessment methods with digital images under foggy conditions. Transport Res Rec. 1862:95-108.

[ILE] Institute for Lighting Engineers. 2005. The outdoor lighting guide. New York (NY): Taylor & Francis. 379 p.

Itti L. 2005. Model of bottom-up attention and saliency. In: Itti L, Rees G, Tsotsos JK. eds. Neurobiology of Attention. London (UK): Elsevier Academic Press. 744 p.

Itti L, Koch C, Niebur E. 2002. A model of saliency-based visual attention for rapid scene analysis. IEEE T Pattern Anal. 20(11): 1254-1259.

Jakel F, Wichmann FA. 2006. Spatial four-alternative forced-choice method is the preferred psychophysical method for naive observers. J Vision. 6(11):1307-1322.

Jay P. 2002. Subjective criteria for lighting design. Lighting Res Technol. 34(2):87-99.

Jenkins SE, Cole BL. 1982. The effect of the density of background elements on the conspicuity of objects. Vision Res. 22(10):1241-1252.

Kimura H, Noguchi T. 2002. Measurement of effective luminance in actual visual field using digital camera. J Architecture (Japan). 551:23-27.

Moberly NJ, Langham MP. 2002. Pedestrian conspicuity at night: failure to observe a biological motion advantage in a high clutter environment. Appl Cognitive Psych. 16(4):477 485.

Moore T, Graves H, Perry MJ, Carter DJ. 2000. Approximate field measurements of surface luminance using a digital camera. Lighting Res Technol. 32(1):1-11.

Nasar JL. ed. 1998. Environmental aesthetics: theory, research and applications. Cambridge (MA): Cambridge University Press. 560 p.

Nothdurft HC. 2000. Salience from feature contrast: variations with texture density. Vision Res. 40(23):3181-3200.

Nothdurft HC. 1992. Feature analysis and the role of similarity in pre-attentive vision. Perception and Psychophysics, 52:355-375.

Nothdurft HC. 1993. Saliency Effects Across Dimensions in Visual Search. Vision Res. 33(5-6): 839-844.

Palmer SE. 1975. The effect of contextual scenes on the identification of objects. Mem Cognition. 3(5):519-526.

Paulmier G, Brusque C, Carta V, Nguyen V. 2001. The influence of visual complexity on the detection of targets investigated by computer generated images. Lighting Res Technol. 33(3): 197-207.

Poulton EC. 1989. Bias in quantifying Judgments. London (UK): Lawrence Erlbaum Associates Publishers. 304 p.

Rosenthal R. ed. 1991. Meta-analytic procedures for social research. Thousand Oaks (CA): Sage Publications, Inc. 168 p.

Shuttleworth S. 1980. The use of photographs as an environmental presentation medium in landscape studies. J Environ Manage. 11:61-76.

Smith PL. 1998. Attention and luminance detection: a quantitative analysis. J Exp Psychol Human. 24(1):105-133.

Treisman A, Gelade G. 1980. A feature-integration theory of attention. Cogn Psychol. 12(1): 97-136.

Turatto M, Galfano G. 2000. Rapid communication: Color, form and luminance capture attention in visual search. Vision Res. 40(13):1639-1643.

Tyrrell RA, Wood JM, Chaparro A, Carberry TP, Chu BS, Marszalek RP. 2009. Seeing pedestrians at night: visual clutter does not mask biological motion. Accident Anal Prev. 41(3):506-512.

Wager JA, Heisler GM. 1986. Rating winter crown density of deciduous trees: a photographic procedure. Landscape J. 5(10):9-18.

Wolfe JM, Oliva A, Horowitz TS, Butcher SJ, Bompas A. 2002. Segmentation of objects from backgrounds in visual search tasks. Vision Res. 42(28):2985-3004.

Navaz Davoudian PhD (1)

(1) The Bartlett School of Graduate Studies, University College London, Central House, 14 Upper Woburn Place, London, UK, WC1H ONN, Email:
Ratio of the occupied parts of
the images by light patterns
in different level of
background density based on
the number of boxes and
pixels occupied

Background    Number     Ratio of    Areas Occupied     Ratio of
Patterns     of Boxes    Number of     by Patterns        Areas
Density                  Boxes (%)      (Pixels)       Occupied to
                                                        the Whole
                                                        Image (%)

              Out of                     Out of
Scene 1      432 boxes                248520 pixels
No               0           0              0               0
Low             91          21            5864              2
Medium          133         30            9245             3.7
High            220         50            12991            5.2
              Out of                     Out of
Scene 2      520 boxes                271038 pixels
No               0           0              0               0
Low             92         17.7           7656             2.8
Medium          140         27            9413             3.4
High            202        38.8           13859            5.1

The results of rating the
Background Density of light
patterns in different levels

Subjects' ID                   Scene 1

                  High   Medium   Low   NO
1                  6       5       3     1
2                  5       3       2     1
3                  6       3       2     1
4                  6       5       2     1
5                  6       5       3     2
6                  5       4       3     1
7                  6       4       2     1
8                  6       4       3     1
9                  6       3       2     1
Mean              5.8     4.0     2.4   1.1
Std. Deviation    0.4     0.9     0.5   0.3

Subjects' ID                   Scene 2

                  High   Medium   Low   NO
1                  6       4       2     1
2                  6       4       1     1
3                  6       4       3     1
4                  6       5       3     1
5                  6       5       4     1
6                  6       5       3     1
7                  6       5       5     1
8                  5       4       3     1
9                  6       5       4     1
Mean              5.9     4.6     3.1    1
Std. Deviation    0.3     0.5     1.2   0.0

Mean, standard deviation and
median of background
luminance in different
density of light patterns

BGD Level   Mean Pixel   Std Dev

Scene 1
High          54.91       59.82
Med           52.98       57.78
Low           51.39       54.08
No            48.77       48.95
Scene 2
High          38.92       56.73
Med           38.64       53.16
Low           38.63       49.54
No            38.60       36.32

Measured luminance of Scene
1 & 2 on the test condition

Luminance       Target       Darkest Areas       Immediate
contrast    (cd/[m.sup.2])   (cd/[m.sup.2])     Background
                                              of the Target *

Scene 1
CL = 0          36.04             0.58             16.44
CL = 3          114.2             0.62             22.81
CL = 5          125.6             0.59              7.9
CL = 10         121.7             0.57              2.4
Scene 2
CL = 0           45.5             0.64             23.61
CL = 3          114.8             0.68             20.1
CL = 5          124.8             0.56             12.24
CL = 10         124.8             0.52              4.5

* Average of 3 points luminance in immediate
surroundings of the target which does not
contained light patterns.

Results from Background
Density effect: Number of
votes (scores) achieved by
images in different levels of
Background Density and
luminance contrast

Background       Luminance contrast
Density                                  Total
              0      3      5      10
No            53    480    662    771    1966
Low           82    418    581    657    1738
Med          170    336    442    604    1552
High         101    313    392    550    1356
Total        406   1547   2077   2582

Friedman test- Results from
comparing the Background
Density Data in different
levels of luminance contrast

Luminance Contrast   Chi square value   p value

0                         31.02          <0.01
3                         162.96         <0.01
5                         391.97         <0.01
10                        324.94         <0.01

Results from Wilcoxon Signed
Ranks Test: significant
testing between the saliency
values in different levels of

  BGD           [C.sub.L] = 0    [C.sub.L] = 3
Low and No        Z =-2.43          Z=-4.11
p value             <0.05            <0.05
Low and Med       Z =-3.37          Z =-5.53
p value             <0.05            <0.05
High and Med      Z =-2.69          Z =-2.98
p value             <0.05            <0.05

  BGD          [C.sub.L] = 5    [C.sub.L] = 10
Low and No        Z =-5.47         Z =-8.48
p value            <0.05             <0.05
Low and Med       Z=-10.14         Z =-7.84
p value            <0.05             <0.05
High and Med      Z =-2.47         Z =-1.06
p value            <0.05             >0.05

Comparisons of Size effect of
BGD where it was significant
on saliency by increasing the
density of light patterns.

  BGD          [C.sub.L]=0   [C.sub.L]=3   [C.sub.L]=5   [C.sub.L]=10
Low and No      r =-0.05      r =-0.09       r=-0.12       r =-0.20
Low and Med     r =-0.07       r=-0.13      r =-0.23       r=-0.18
High and Med    r =-0.06      r =-0.07      r =-0.05         N.S

Results from Correlation Test
between Saliency Value,
Background Density and
Luminance Contrast

Correlation test                          r      [r.sup.2]   p value

Between Saliency Value and BGD         -0.511      0.261     <0.01
Between Saliency Value and CL           0.599      0.358     <0.01
Between Saliency Value and BGD while
  CL is controlled                     -0.631      0.398     <0.01
Between Saliency Value and CL while
  BGD is controlled                     0.835      0.697     <0.01
Between Saliency Value and BGD while
  CL= 0                                 0.241      0.058     <0.01
COPYRIGHT 2011 Illuminating Engineering Society
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2011 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Davoudian, Navaz
Article Type:Report
Geographic Code:4EUUK
Date:Oct 1, 2011
Previous Article:Tangible lighting controls--reporting end-users' interactions with lighting control interfaces.
Next Article:New wine in old skins.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |