.

Determining Visibility: Contrast is Fundamental

Marc Green

Contrast detection is the basic task from which all other visual behaviors are derived.
- Illuminating Engineering Handbook.


This article outlines the methods required for a scientific visibility analysis. It dispels common misconceptions about visibility and briefly describes the procedures for determining whether the viewer could potentially have seen an object.

Usually, the best method compares viewer contrast sensitivity to the existing physical contrast, the difference between object and background luminances. If the physical contrast is greater than the threshold, then the object is "visible." However, this does not mean that the viewer will consciously perceive the object. Visibility is concerned only with sensing and whether it is within the realm of possibility to see perceive the object. Perception requires the engagement of attention, which is a related, but separate visual function. The topic is highly technical, so I have simplified the more esoteric aspects. For more details, see Green (2024).

I also discuss two other issues. The first is alternative methods for specifying visibility. Although contrast detection is the best and most general, other methods may be applicable in special situations. Second, I explain why photographs should not be used to assess visibility and briefly highlight the important difference between visibility and conspicuity.

General Background

Intuitively, visibility is the ease with which a viewer can detect an object. For scientific purposes, however, visibility requires a more precise definition: most often the relationship between luminance contrast and threshold3.

Luminance is the amount of light reaching the eye from a given direction of space and roughly correlates with the experience of brightness. Figure 1 shows that viewers usually see light that was emitted by a source and then reflected off a surface to the eye. For example, car headlamp beams (source) illuminate a pedestrian (surface) down the road. The amount of light falling on the pedestrian surface is the illuminance. However, only some fraction of this light reflects back to the viewer. The reflected portion is the luminance and is what ultimately matters in visibility.


Figure 1 Illuminance is the light falling on a source. Luminance is the amount of light reflected from the surface to the eye.

Luminance depends on the surface reflectance. Black pedestrian clothing, for example, has low luminance because it reflects very little (roughly 2-5%) light back to the driver's eye. White clothing has higher luminance because it reflects most (roughly 60-80%) of the incident light. This is the reason that pedestrians in white clothing will obviously be much more visible at night, at least when the background is dark.

However, visibility depends, not on luminance, but rather on physical contrast, the luminance difference between an object and its background. For example, a piece of coal seen against a sheet of white paper is visible even in dim light because the contrast is high. A piece of white paper seen against snow has little visibility even in bright light because the contrast is low.

Threshold2is the minimum amount of physical contrast necessary for seeing. Viewers with lower contrast thresholds are said to have higher sensitivity. If the object contrast has exceeded threshold, then it is visible, and the viewer could "in principle" see it. Note the qualifier "in principle" because conscious perception involves much more than visibility.

Procedures For Determining Visibility

1. Recreate the scene

In an ideal world, the initial step is to determine the amount of contrast available through an exact scene re-creation. An ideal re-creation uses exactly the same object under exactly the same lighting conditions. Perfect re-creation is sometimes impossible, but good approximations create minimal error. For example, the original objects (car, clothing, etc) may be unavailable, but carefully chosen exemplars usually suffice.

It is sometimes more difficult to create the same lighting conditions. For a nighttime roadway collision, the main light sources are usually overhead road lighting, sun/sky and/or the approaching car's headlamps. The object should be located at the same position relative to road lighting as in the original event. The sun's sky position, its altitude and azimuth (clockwise direction), may also be critical. Measurements made on the anniversary date and time of the original accident ensure that sun position is correct. The best way to re-create car headlamps is to place an exemplar vehicle at various distances from the target object and to measure the luminance of both the target and the background. Alternatively, research data provide the contribution of headlights.

If the anniversary date is not feasible, then measurements can still be made on another day with time correction. To correct for altitude, consult astronomical tables that give the sun's altitude at accident's time and place. The tables will also provide the time when the sun is at the same altitude, and measurements should be made, for every other day of the year. These tables are only approximations, however, as landscape features such as hills can affect the sun's position on the horizon. Further, atmospheric refraction can alter apparent sun altitude. Moreover, there is no way to correct for variations in the sun's azimuth, which will matter most on clear days when the sun is near the horizon and the viewers sightline. Lastly, if the accident occurred at full night darkness (about 1-1.5 hours before sunrise or after sunset) then there is no need to worry about sun position or exact times. The only caveat is that man-made light sources not differ. At 9:00 PM there may be lights from buildings, traffic, etc. that were not present at 2:00 AM.

Moonlight is usually not important. Even a full moon high in the sky on a clear night contributes very little light, about 0.3 lux. It is only a small factor in extraordinarily dark conditions and with high reflectance objects. Conversely, weather conditions are far more important. Rain, snow, fog and even variations in cloud cover can affect luminance measurements and visibility assessment. Proximity to urban areas can be important. Even sky glow Sky glow from city lights produce more illumination, especially on nights when clouds much of the light back to the ground

2. Determine the luminance and contrast of the target and background

Next, determine luminance of object and background. The best and most direct method uses a specialized light meter, a luminance photometer, which measures the light reaching the eye. Ideally, the meter is located at the original viewer's eye position and aimed along his sightline so that it "sees" what he saw. This is important because different viewing positions can alter the background immediately surrounding the object, resulting in different contrast. The user takes readings by aiming the meter at the object and at background points immediately adjacent to the object. Lastly, if the viewer looked through a transparent surface such as a windshield, the measurements should be taken through the surface or adjusted downward to take transmission into account.

Direct luminance measurement is sometimes difficult or impossible for various reasons. The scene may have changed substantially since the accident occurred - roadways are re-designed, extra lighting is added, old bulbs replaced with new, etc. At other times, it is impossible to re-create the scene safely, as when a collision occurred on a busy freeway. In these cases, luminance can be calculated from data on illuminance and reflectance, since luminance is equal to illuminance time reflectance.

Both illuminance and reflectance may themselves either be directly measured and/or estimated from tables and other reference material. At the extreme, all the required illuminance and reflection data may be estimated from standard tables, so that no scene re-creation is necessary. However, every step away from direct measurement introduces uncertainty and requires more assumptions.

3. Calculate the contrast

Once luminance values are known, contrast calculation is straightforward. Several calculations are possible, but the most common is the Weber fraction:



where

Cth is the contrast threshold. Contrast sensitivity is 1/Cth.
LO is the object luminance.
LB is the background luminance.
Lv is the equivalent veiling luminance.

The Lv is added when a disability glare source is present and producing illumination at the eye. It is the luminance of a uniform area of light that would change contrast sensitivity by the same amount as the glare source. Methods for calculation Lv are discussed in Green (2024).

4. Determine contrast threshold

Determining the contrast threshold is by far the most difficult step in the visibility analysis. There is no single number because at least 15-20 variables can significantly affect contrast thresholds. The most important include, background luminance, size, viewer age, location and viewing duration. No standard chart or table exists that allows easy threshold lookup for a given situation. However, a good starting point is use of a "visibility model," which combines data from many different sources into a set of mathematical equations. The most popular set is the "Adrian model" which is incorporated in the ANSI/IESNA RP-8 "small target visibility model" (STV) that in used in designing roadway lighting. This Adrian model is a long and complex set of equations that can be entered into in a spreadsheet or other mathematical computer program.

The only alternative with much traction is DETECT, which is a much simpler model that was developed at Ford motor company. It differs from Adrian in its data source and models a slightly different set of visual variables. Both start by calculating thresholds given the two most important variables, object size and background luminance. Adrian then adds corrections for viewing duration, viewer age and polarity. DETECT also corrects for age but does not have either polarity or viewing duration as a variable. The age corrections in the models differ significantly. The DETECT function is probably more accurate in low lighting (Green, 2024).

Both models have limitations. They fail to incorporate many important variables, the most obvious being retinal eccentricity. Threshold increases greatly in peripheral vision, especially for smallish objects. As a result, the models provide only foveal thresholds, as they do not adjust for retinal eccentricity. It may be possible, however, to add a correction using the "cortical magnification factor" (Green, 2024).

5. Compare physical contrast to threshold contrast

The final step compares the physical contrast to the threshold. If the physical contrast is higher than the threshold, then the object was visible. Conversely, if the physical contrast was lower than the threshold, then the object was not visible.

However, the model threshold cannot be directly used because it is based on 50% detection data from laboratory subjects. The thresholds must be increased using two multipliers. First, the 50% detection level is too low for a "commonsense" seeing level. Both Adrian and DETECT have multipliers that increase threshold to a higher detection level. Adrian uses 2.6 to achieve a 99+ percent detection level while DETECT has a choice of multipliers for different detection levels.

Second, research subjects have many advantages over real viewers in the real-world. The laboratory subjects have little uncertainty; they know exactly what to look for, exactly when it will appear, and exactly where it will be. They have a simple, uniform background. They get to practice over and over, etc. Real viewers face a much more uncertain and complex world that produces significantly higher thresholds. As a result, the threshold value obtained from the model must be adjusted upward by a field factor or visibility level (VL), a multiplier which raises laboratory-based thresholds up to a more realistic value. The best field factor varies across circumstances, but in the Adrian model, research suggests that 10 is realistic field factor in many situations. In sum, the physical contrast must be 10 times the model threshold for real-world visibility.

The introduction of field factors reflects the differences between research and real viewers in both sensory factors (viewing duration, retinal eccentricity, etc.) but also in cognition (expectation, recognition). However, viewer may still to fail to see even a highly visible object, the phenomenon of inattentional blindness, when cognitive factors are strong. "Habit is the sixth sense which overrides the other five."

Other Ways Of Determining Visibility

There are several alternatives to contrast threshold for determining visibility under special conditions. These include:

  • Visual acuity: the smallest resolvable detail. Acuity comes is several varieties, but the most commonly used is "resolution acuity". Its major virtue is simplicity, as it easy to measure and results in a single number. Its downside is that it does not reflect the more common real-world task of seeing contrast in objects of all size. These days, it is seldom used except in a few specialized situations when speed and simplicity are the overriding concerns.

  • Illuminance criterion: the amount of illumination falling on an object. Meaurement is simple and cheap and requires little expertise. It is used when the values of other variables cannot be specified. This simplicity makes it highly unreliable for many reasons. Most obviously, it ignores object reflectance and background luminance. It also fails to consider many other critical variables such as size, viewer age, retinal location etc.

  • Visual range: the distance at which an object can be seen. It is most commonly used when a viewer must detect a small, negative contrast object at a long distance. In such cases, atmospheric effects of extinction and weather affect visibility and become variables. Size is ignored, but the method ultimately depends on an assumed contrast threshold, usually between .01 and .05. If the threshold is set to .03, then the result is Middleton's "meteorological" range.

  • Illuminance at the eye: the amount of light needed at the eye to see an object at a given distance. This is most commonly used when the viewer must detect a distant, small, positive contrast object, such as a signal light at sea. The object is assumed to be a point source so size is not a variable. The illumination to detect an object at a given distance is typically calculated using the International Association of Lighthouse Authority (IALA) model, which incorporates an extinction factor. It also suggests different values based on background lighting and clutter.

Visibility Analysis Vs. Photography

There common aphorism says that "there is always a well-known solution to every human problem-neat, plausible, and wrong."4 The use of photographs to assess visibility is a perfect example. A photographer snaps scene pictures, which are later shown to viewers, who then judge visibility from the image. This will not work. It is not a "visibility anslysis. It is merely photography. I have explained why elsewhere, but briefly the major reasons include:

  • Cameras do not accurately capture light levels in many scenes. Photographs, videos photographs and electronic displays do not reproduce them accurately. Factors such as field size and depth of field are usually incorrect.

  • The person inspecting the picture is like the research subject in knowing exactly where to look, exactly what to look for and exactly what is going to happen next. There is no stress or urgency.

  • The person inspecting the picture can look long as desired. This is a very different set of viewing conditions from a driver who might have to make a split second decision about an object seen out of the corner of his eye.

  • Perception is tied to action. People who are not performing the same task will not have the same perceptions. First, attentional allocation depends on task performance. Viewers have their attentional deployment automatically controlled by task demands. Second, locomotor tasks such as driving are largely controlled by the ambient visual system that operates outside of awareness and not under volitional control. The person inspecting the picture does not have this part of the brain engaged. Lastly, action affects perception. Cognition is partly embodied - tied to physical actions. People performing dissimilar tasks literally see the world differently.

  • As a rule,photographs should not be used to assess visibility or perception.

    Conclusion

    Visibility analysis determines whether an object had sufficient contrast to be seen. It compares the luminance contrast of the object against its background to the threshold contrast for the existing circumstances. The ideal procedure requires an accurately re-created scene, good light measurement instruments, and extensive training and background in psychophysics, the scientific discipline that studies visibility. Without an accurate re-creation, it is still often possible to make a visibility analysis by using reasonable assumptions.

    However, many complications can arise. The background may not be uniform so that there are different contrasts at different edge locations. The field factor is sometimes difficult to estimate. With small objects, it may be difficult to measure the object and background luminances separately. Etc, etc.

    I have already noted that visibility reveals only whether a viewer can theoretically see an object. However, a viewer will consciously see an object, regardless of visibility, only if it has high conspicuity, meaning that it is good at engaging attention. While increased visibility can enhance conspicuity, most objects, even those that are highly visible, go unseen because humans have limited attentional resources and can only become aware of a small part of the world at a given moment.

    Many sensory and cognitive factors determine conspicuity. Most people intuitively assume that conspicuity depends on sensory factors such as higher contrast, motion and color. While these can be important, research shows that conspicuity often depends more on cognitive factors, such as expectation and goals. Viewers learn to tune their attention to relevant objects and locations where they expect to find needed information for the task at hand. Remaining visual information is filtered away and is not consciously perceived. In sum, visibility analysis only determines whether seeing is possible and not whether seeing is likely. To determine whether a normal viewer is likely to see some object or piece of information, visibility analysis is the starting point and not the ending point.

    Endnotes
    1The title is an inside joke: a play-on-words for all the "gratingologists" out there.

    2Luminance contrast is also sometimes called "brightness contrast". Although visual scientists sometimes say this in casual speech, they are aware that it is an error. Brightness is the psychologically experienced quantity of light, as opposed to luminance, the physical quantity of light. Luminance is only one of many factors that determine brightness. Also, luminance is usually the most significant type of contrast for visibility, but others, including color, motion, and texture, may be important in rare cases.

    3Technically, threshold is a percent seen value on a psychometric function. It is usually 50% in Yes/No methods, but it can be some other value depending on the measurement paradigm. As such, it has some degree of arbitrariness. Some analyses, most notably Signal Detection Theory, do not even employ the concept of threshold.

    4Mencken, H. 1917. The Divine Afflatus. New York Evening Mail, 16 November.