What do Purves and Lotto's Illusions Actually Show?

11.8 What is a Colour? Perception or Property?


[0:00]. This is the first of a series of videos on current scientific understanding of the most fundamental aspects of colour, especially the basic attributes or dimensions of colour and their physical and biological basis. Individuals have many different opinions about colour, but in these videos my main aim is to explain and illustrate the consensus of opinions embodied in the terminology of the International Lighting Vocabulary published by the Commission Internationale de L’Eclairage or CIE for short.

[0:33] The CIE was founded in 1913 and is the organization responsible for the international coordination of lighting-related technical standards. In relation to colour the CIE established the framework of modern colorimetry or colour measurement in the first decades of the 20th century, various colour spaces that form the basis of colour management and colour technology, including CIE L*a*b* familiar as the Lab space in Photoshop, a series of standardized light source specifications such as CIE Illuminant D50, D65 and so on, various formulae used in industry for quantifying colour differences, and ongoing research into mathematical models to predict the colour appearance of a given stimulus under different conditions. The CIE International Lighting Vocabulary or ILV, first published in 1938 and revised periodically since then, gives definitions for more than 1400 terms and is by far the most comprehensive and authoritative source on the scientific terminology of light and colour. Although previously rather expensive and limited in distribution, the ILV is now freely available online. But although finely honed by colour scientists over many decades, the purely verbal definitions and accompanying brief notes and formulae in the ILV are often difficult to understand for non-specialists, and the explanations and illustrations in these videos are hoped to help remedy this.  

These videos will draw on material covered on my website The Dimensions of Colour and in my chapter on colour spaces in the forthcoming Routledge Handbook of Philosophy of Colour.

[2:20] In this first video I’ll look at the most fundamental questions of all: “What is a colour?” and “In what sense can the colours of lights and objects be measured?”.

The Scientific View of Colour

When I perceive a light or an object to be red or white or green I have the impression that this colour is a property of the light or the object, but what is not immediately clear is whether this red or white or green colour is a property of the light or object that my visual system directly detects, or whether it is the way in which I perceive a property of the light or object, a way of perceiving that property that my visual system creates?

While the first alternative has been called by some philosophers the “common sense” view of colour, scientific consensus is firmly with the second alternative.

[3:13] Elements of this scientific view of colour can be traced back to antiquity via Descartes and Galileo but began in substantial detail with Sir Isaac Newton's researches into the physical basis of colour.

In this well-known passage from his Opticks of 1704 Newton said “… And if at any time I speak of light and rays as coloured or endued with Colours, I would be understood to speak not philosophically and properly, but grossly, and accordingly to such conceptions as vulgar People in seeing all these Experiments would be apt to frame. For the rays to speak properly are not coloured. In them there is nothing else than a certain power and disposition to stir up a sensation of this or that Colour.”

[4:05]. While not the first to make a claim of this sort, Newton added strong supporting evidence by showing that the colour of an isolated light can be predicted from the overall balance in a circular diagram of its spectral components. The different-sized circles labelled p, q, r and so on here represent, in Newton’s terms, unequal “numbers of rays” from each sector having their “common center of gravity” at point Z. We would now call these “rays” different wavelengths of light and call this distribution of energy through the spectrum a spectral power distribution, that is, a spectral distribution recorded in terms of the energy of the component wavelengths.  

[4:50] Newton found that the direction of imbalance in the circle of the “common center of gravity” (towards Y here) determines the hue of the light (such as red or green, or in this case orange), and the amount of imbalance (the distance OZ here) is proportional to, in Newton’s terms, “the fulness or intenseness of the colour, that is, to its distance from whiteness”. We now call this purity of colour of a light saturation.

The crucial point here is that whitish orange as a colour of light is the way in which we perceive a moderate imbalance among the light’s component wavelengths towards a certain part of the spectrum. Although this whitish orange colour seems to us to be a property residing in the light, it exists in the light itself only as the “power and disposition” of this balance of spectral components to evoke this colour perception in the observer.

[5:50]. Now an isolated wavelength of the spectrum has the greatest imbalance possible in a given direction, and appears, Newton says, “intense and florid in the highest degree”, that is, at the maximum saturation possible for that hue. Spectral orange, like whitish orange, is the way in which we perceive a specific imbalance of spectral components. In saying that the rays to speak properly are not coloured, Newton asserts that this spectral orange colour, though it seems to us to reside in a “ray” or wavelength of light, exists in that light only as the disposition of this spectrally imbalanced light to evoke this perception in the observer. Tempting though it is to think or say so, there is no reason to suppose that the spectral orange colour in our awareness is also located in the wavelength of light itself, or is located anywhere when a mixture of this wavelength with other wavelengths evokes a different colour perception.

[6:55] Of course it’s very awkward to carefully speak of “orange-making” or “orange-evoking” wavelengths instead of just orange wavelengths, and you’ll notice that here as in many other places Newton talks as if the rays actually are coloured when he speaks of “all the colours in the given mixture”. But Newton had made it very clear that when he speaks this way in the Opticks he is speaking “not philosophically and properly, but grossly”.

[7:23] When the wavelengths of the spectrum are present in a similar balance to daylight, their “common center of gravity” is at the centre of the circle, so that the light lacks hue and saturation, appearing “white” or achromatic. To speak properly, such a light does not contain all the hues of the spectrum (as many science educators still say and as my multicoloured spectral distribution rather misleadingly suggests). It contains wavelengths that when viewed in isolation have dispositions to evoke these hue perceptions, but which combined in this balance have a disposition to evoke no hue.

[8:06] Newton's centre-of-gravity principle implies that most colours of light can be evoked by many different combinations of wavelengths having the same centre of gravity. This whitish orange light is an unequal mixture of wavelengths from all parts of the spectrum, but we should be able to evoke an exactly matching colour with an unequal mixture of just two wavelengths that appear orange and cyan blue in isolation, or just two wavelengths that appear orangish red and bluish green in isolation, and there are innumerable other possible combinations of two, three or more wavelengths.

A colour perception of whitish orange thus tells us nothing about the specific wavelengths present in a light, only the overall balance of its spectral power distribution. Colour vision is not wavelength detection and specific wavelength composition is not a perceivable property of light.

Some science communicators, especially from a physics background, are inclined to say that our visual system is “fooled” or “tricked” when we perceive different mixtures of wavelengths as the same colour. This assumes that our visual system was supposed to detect and identify the individual wavelengths present and failed. But who says that our visual system is supposed to do this? Even approximately monochromatic light is quite uncommon in nature. Detecting individual wavelengths of light, so important for a physicist, is unimportant for survival.

[9:44] Newton showed that what we perceive as the colour of an isolated light is the overall balance of its spectral power distribution considered as a two-dimensional system. We now understand how our vision detects variations in this balance and can see that this two-dimensional character is not a physical property of light but is a product of the workings of our visual system.  This system is called trichromatic because it involves three receptor types called L, M and S cone cells; these respond to all, all but the longest and the short wavelengths of light respectively, and their outputs are compared with each other by a process called cone opponency. This system allows us to distinguish a two-dimensional circuit of directions of imbalance towards long-, middle-, short-, and long- and short-wavelengths. (In contrast a dichromatic species with just two receptor types can only distinguish two directions of imbalance, towards long or short wavelengths)*. Opponent processing also explains why an even balance of the long middle and short wavelengths is perceived as an absence of hue, rather than as three simultaneous hue sensations as we might expect from the way our hearing works. Our visual system does not detect hue and saturation in light; these do not exist as such in light. Our visual system detects variations in the balance of the long-, middle- and short wavelength components of light, and we perceive these variations as variations in hue and saturation.

[11:25]. When different spectral distributions match in colour, for example the various whitish orange lights we considered previously, they do so because they have the same balance of long-, middle- and short-wavelength components from the point of view of the human visual system. These graphs show the spectral power distributions of five matching “white” lights, the smooth distribution of a CIE daylight illuminant called D50, the spiky distributions of two matching CIE fluorescent illuminants, and examples of distributions of CRT and LED screens adjusted to match D50. Despite their finer grained differences, all these lights match in appearance because they effectively have the same overall balance of the three wavelength components. Such stimuli are called metameric, and we’ll see a little later that a set of the tristimulus values, mentioned in this definition, is a measure of this overall balance.

[12:26] So far we’ve been considering colours of lights, but what about colours of objects? Newton additionally recognised that "Colours in the Object” are the object's "disposition to reflect this or that sort of rays more copiously than the rest", or what we would now call the object’s intrinsic spectral reflectance. Newton observed the spectral reflectances of various artists’ pigments directly by shining a solar spectrum on to them in a darkened room.

A colour that we perceive as belonging to an object is called an object colour. An object colour is the is the way in which we perceive the object’s intrinsic spectral reflectance if the object is opaque, or its spectral transmission if the object is transparent. (For brevity I’ll just say reflectance from here on). Once again, we now know that the perceivable property is not the spectral reflectance in all its detail but its overall composition in terms of its long-, middle- and short-wavelength components.

[13:31]. When we can freely examine an object in daylight the object colour we perceive it to have is a very good indication of the overall long-, middle- and short-wavelength components of its spectral reflectance. This remains true to a point under coloured lighting, despite the different spectral distribution of the light the object reflects, due to the capacity of our visual system for a considerable degree of object-colour constancy.

In favourable viewing conditions, white as an object colour is the way in which we perceive a spectral reflectance whose long-, middle- and short-wavelength components are about equal and all approach the maximum possible. Black is the way in which we perceive a spectral reflectance in which these components are all very low. Various blue object colours are the ways in which we perceive various spectral reflectances that have an overall bias towards their short wavelength component.

Perceived Colour and Psychophysical Colour

[14:33] So things like red and green and white are the ways in which we perceive a property of the spectral composition of a light or the intrinsic spectral reflectance of an object. This leads to the 64-million-word question: does the word “colour” properly refer to the perception, in this case white, or to the property that evokes that perception, in this case an even balance of the long-, middle- and short-wavelength components? Many scientists and some philosophers equate the concept of “colour” with the perception, leading them to say simply that “colours do not exist”, meaning that they do not exist physically in lights and objects but only in the mind of the perceiver. Some other philosophers equate “colour” with the property of the spectral distribution of a light or reflectance of an object that disposes it to appear red, white etc, or even with individual physical spectral distributions or reflectances.

[15:37] The CIE International Lighting Vocabulary acknowledges two senses of the word colour and provides formal definitions of both:

Perceived colour is defined as the "characteristic of visual perception that can be described by attributes of hue, brightness (or lightness) and colourfulness (or saturation or chroma)". Note that in CIE terminology hue, brightness, lightness, colourfulness, saturation and chroma are all defined as attributes of visual perception, not physical properties of lights or objects.  White and various reds and greens are examples of perceived colours.

A psychophysical colour is defined as a "specification of a colour stimulus in terms of operationally defined values, such as 3 tristimulus values". And tristimulus values are in turn defined as the "amounts of the 3 reference colour stimuli, in a given trichromatic system, required to match the colour of the stimulus considered”.

A psychophysical colour is the measurable property that we use in colorimetry to specify the colour of a light or an object. It’s called psychophysical because it is not a purely physical property but is partly shaped by the visual system of the observer.

[17:00] To understand the concepts of psychophysical colour and tristimulus values, consider your computer screen, which is after all a device for generating light with varying long-wavelength (or “R”), middle-wavelength (or “G”) and short-wavelength (or “B”) components. Imagine viewing your computer screen alongside various areas of the room you are in, and matching the light coming from each area by adjusting amounts of the R, G and B lights on your screen. If your screen was dark enough and you could turn up the lights high enough you could in this way specify the colour of the light from most areas you see using a set of tristimulus values consisting of the amounts of the R, G and B lights on your screen that matched it. You could then plot these amounts as a three-dimensional, cubic colour space – just as your digital camera does for you in its own way automatically.

[18:05] You could also plot the ratios of these three tristimulus values as a two-dimensional diagram, a triangle. This triangle would show the balance or ratios of the long-, middle- and short-wavelength components independent of the total amount of light. Most lights you see could be plotted in these ways, apart from some highly saturated lights, including the lights of the spectrum, that would lie outside the cube and outside the triangle.

[18:37] A chromaticity diagram is a two-dimensional graph that, like our RGB triangle, shows the ratios of three tristimulus values independent of the total amount of light. The CIE 1931 xy chromaticity diagram is not the latest but is still the most familiar descendant of Newton’s colour circle. You might possibly have encountered one of these diagrams in a technical review comparing the range of colours obtainable on different RGB screens, or contained in different RGB colour spaces, as different triangles. The xy chromaticity diagram and a three-dimensional colour space CIE XYZ use three purely theoretical “primaries”, X, Y and Z, that lie outside the range of actual lights so that their proportions can specify all colours of lights. In the xy diagram the colours of the lights of the spectrum follow a horseshoe shaped line, and mixtures of the extreme ends of the spectrum follow a straight line called the line of purples. All actual lights can be matched to points in the area enclosed by these lines and can therefore be specified in terms of proportions of X, Y and Z components.

The three tristimulus values do not correspond directly to the amounts of the three wavelength components in a light, but it’s nevertheless true that lights with high X values are high in long wavelengths, lights with high Y values are high in middle wavelengths and lights with high Z values are high in short wavelengths, and that an isolated light for which X, Y and Z are about equal will appear white to most observers. As colour vision varies between different observers, calculation of CIE X, Y and Z involves a mathematically defined “standard” human observer.

[20:37] Each point on the xy diagram represents a particular ratio of the X, Y and Z tristimulus values and specifies a particular balance of long-, middle- and short-wavelength components as perceived by the “standard” human observer. Lights having the same XYZ values would match exactly to this observer - although the perceived colour of these two matching or metameric lights would be influenced by factors including the surrounding areas.

[21:10] The CIE term for Newton’s concept of the direction of imbalance in the “common centre of gravity” of the "numbers of rays" (the direction OY in his circle) is dominant wavelength. Dominant wavelength is the measurable psychophysical property related to the perceived colour attribute of hue.

A common misconception is that hue is an intrinsically linear scale corresponding to the wavelengths of the spectrum and is only arbitrarily bent and joined at its ends to form a hue circle or “colour wheel”. But hue is our perception not of wavelength but of dominant wavelength, which has a 360 degree range to include directions of imbalance towards the long-, middle, short- and long and short wavelengths of the spectrum.

[22:02] The CIE term for Newton’s concept of the amount of imbalance in the “common centre of gravity” of the "numbers of rays" (the distance OZ in his circle) is purity, which can be formulated in a couple of different ways as colorimetric purity and excitation purity. Purity is the measurable psychophysical property related to the perceived colour attribute of saturation.

[22:29] We’ll see in a later video that there are also psychophysical measures called luminance and luminous reflectance that relate to the perceived colour attributes of brightness and lightness (or greyscale value). Colours can be specified psychophysically in various ways including tristimulus values (such as CIE XYZ), as sets of three psychophysical measures such as dominant wavelength, purity and luminance, or by a combination of chromaticity and luminance using a space called CIE xyY. CIE xyY can be used to specify the colour of an object in terms of the xy chromaticity and the relative luminance or amount of light that the object reflects under a specified daylight illuminant, and can also be used to represent digital colour spaces.

In Newton’s terms, perceived colour is colour “in the Sensorium” or mind, while psychophysical colour specifies colour “in the rays” or “in the object”. Two lights having the same psychophysical colour specification have the same “power and disposition to stir up a sensation of this or that colour" and so they match in colour when viewed in the same conditions. However, the perceived colour of these matching lights can vary depending on those conditions. A psychophysical colour specification therefore does not equate to a unique “true” perceived colour and can only be converted to a set of perceived colour attributes if the viewing conditions are specified.

[24:15] It will help to understand the difference between perceived colour and a psychophysical colour if we consider contrast phenomena.  Contrast phenomena demonstrate that our colour perceptions do not depend entirely on the spectral distribution of a stimulus but are influenced by surrounding areas. Contrast phenomena are disconcerting to the so-called “common sense” view of colour, that colours are properties of lights and objects that our visual system simply detects. In this example my visual system creates two different perceived colours from squares that match physically and have the same psychophysical colour specification (R 180 G 108 B 108, one of the 16.7 million colour specifications possible on my screen).

[25:01] Science necessarily distinguishes between colour as a perception and the measurable property of a stimulus that we use to specify its colour, because the perceived colour of a given stimulus can vary not only with the surrounding areas, but also with such factors as lighting, how long the observer views the stimulus and what they have viewed previously, and the individual characteristics of the visual system and even the attitude of the observer.

If you concentrate your attention alternately first on the vertical and then on the horizontal elements of this pattern, you may notice that the perceived colours that your visual system creates alternate, between vertical reddish and greenish stripes and horizontal bluish and olive stripes.

[25:53] Now stare steadily at the black dot in the centre of the pattern. If you’re able to keep your eyes very still for long enough you may notice the perceived colours starting to fade to grey, at least momentarily. Your perception of the amount of imbalance diminishes due to adaptation. In any case if you’ve been focussing steadily on the black dot ....

[26:18] you’re probably now perceiving faint coloured afterimages. And you may again discover that you can change the afterimage colours you perceive by concentrating alternately on the vertical and horizontal elements. When we see coloured afterimages, we have a perception of an imbalance that does not exist physically in the spectral distribution of the stimulus, again due to adaptation.

As you continue to look and adapt to the physically uniform area these afterimage colours gradually fade and your visual system will eventually create a uniform white perceived colour. But take this screen into a candlelit room and the perceived colour might be distinctly bluish; take it outside into sunlight and the perceived colour might be a dark grey.


[27:10]. To sum up, when we perceive a light or an object to be red, or white or green in colour, we have the impression that this colour is an intrinsic property residing in the light or object. And for many purposes it’s perfectly reasonable to speak of the light or object as being red or white or green coloured. As Newton found, this is a lot more convenient than constantly speaking of “red-making” and “green-making” wavelengths and so on. We should bear in mind however that this is speaking “not philosophically and properly, but grossly”, and is potentially misleading for two reasons.

[27:47] Firstly, scientific consensus is that the red or white or green colour does not reside in the light or object but is the way in which we perceive the spectral composition of the light or the spectral reflectance of the object. Red and green are the ways in which we perceive an overall imbalance towards different parts of the spectrum. White as a colour of light is the way in which we perceive an overall even balance of spectral composition. White as an object colour is normally the way in which we perceive a spectral reflectance that is both evenly balanced and very high.

[28:28] Note though that the perceivable property of a spectral distribution is not its precise physical wavelength composition but is a psychophysical property, shaped in part by the visual system of the observer. For a trichromatic observer this psychophysical property perceived as colour relates to the overall amounts of the long-, middle- and short-wavelength components of the spectral distribution. The familar two-dimensional circuit of hues is a product of the trichromatic visual system and does not exist physically in light.

[29:04] Secondly, the perceived colour of a given light or object is not as fixed as it may seem and is influenced by factors relating to the viewing environment and the individual. When we “measure” the colour of a light or object using colorimetry we are using the word “colour” in a specific sense based on colour matching (psychophysical colour) that captures the disposition of the light or object to appear a certain colour to a mathematically defined “standard” human observer.

A psychophysical colour specification in terms of CIE XYZ or sRGB tristimulus values tells us what the stimulus will match; it does not equate to a unique “true” perceived colour and can only be converted to a set of perceived colour attributes if the viewing conditions are specified. For example, it can be converted to a set of perceived colour attributes in the Munsell or NCS system using tables that assume specified viewing conditions, or by using a colour appearance model to predict its appearance under various other viewing conditions.

[30:14] In the next video we’ll look at what is and isn’t known about how our visual system generates perceptions of hue and colour intensity, and see how the so-called “common sense” assumption that hues like red and blue reside and mix in paints, inks and lights has led to various beliefs about supposedly fundamental components of colour called “primary colours”.

1 Colour perception for a dichromatic individual of a species with trichromatic neural processing (e.g. human dichromats) may be more complex (see Broackes, J. 2010. What Do the Color Blind See?, in J. Cohen and M. Matthen (eds.), Color Ontology and Color Science. Cambridge, MA: MIT Press, 291–405.

Page and video published February 14, 2020. First draft of video published January 3 , 2020.