Colour
By Bob Newman, first published December 2013
One of the most prized properties of a camera is its colour rendition. Photographers will prize the colour rendition of one particular brand or model over another. When discussing these matters phrases such as ‘wonderful skin tones’ and ‘beautiful greens’ abound.
What is ‘colour’
Many people assume that colour is an absolute thing. At school we are taught about Isaac Newton’s experiments with prisms, splitting white light into a spectrum running from red to violet. What tends to be taken away from these lessons is that colour is a matter of wavelength
Human colour vision
The language of the Himba people of Northern Namibia in Africa recognizes five different colours: senandu, which most Asians and Europeans would say cover reds and purples; dumbu, which covers reds, greens and yellows; burou, covering greens and blues; zoozu, covering greens, blues and purples and finally ‘vapa’, white and grey.
Figure 1. The Himba people recognize different colours to most of the other people in the world, indicating that colour is a matter more of human perception than absolute physics.
Clearly Himba people are categorizing colours completely differently to most people. In particular, colours that others would say a ‘green’ may belong to any one of three different colours according to the Himba. There is an open research question as to how much of the Himba’s unique colour perception is to do with language and culture and how much is to do with physically different colour vision. That question will not be addressed here, rather we will consider why such a thing is possible, if colour is simply a matter of wavelength. If this were the case, colours of similar wavelengths would always be categorised together. The reason that it is not is because colour is not purely a matter of wavelength, it is a matter of the balance of response of three different ‘bags’ of wavelength, or ‘stimuli’. Even that is a simplification, because the human brain has at least two different mechanisms of colour classification working simultaneously.
Human colour sensors
The human eye has four different ‘pixel’ types. ‘Rods’ are luminance sensors and found in peripheral areas of the eye. They don’t contribute to colour vision. The colour receptors are ‘cones’. Contrary to what is frequently stated, these are not ‘red’, ‘green’ and ‘blue’ sensors. Conventionally they are classified by their peak wavelengths; ‘S’ for ‘short’ - covering single wavelength light from violet to blue-green, ‘medium’ – covering blue to red and ‘long’, covering blue/green to deep red. The responses to different wavelength light are shown in Figure 2.
Figure 2 (a) The human eye’s colour receptors recognize three colour bands, denoted by ‘S’, ‘M’ and “L’.
Figure 2 (b) A bird’s eye distinguishes four colour bands, more evenly spaced than a human’s.
Note in particular that the ‘M’ and ‘L’ response is quite different. Each photon of light that hits the eyes has a single unchangeable wavelength, which will be registered if it is somewhere between 400 and 700 nanometres in wavelength. Each photon can only be sorted into only one of the S, M or L ‘bags’, since it is a quantum of energy – it cannot be subdivided. The responses of the S, M and L cones indicate how likely an individual photon of a given wavelength is to be sorted into each of the ‘S’, ‘M’ or ‘L’ ‘bags’. For instance, a photon of 440 nanometres wavelength is very likely to be sorted into the ‘S’ bag and unlikely to go into the ‘M’ or ‘L’ bags’, a photon of 600 nanometres wavelength is about three times as likely to end up in the L bag as it is in the M bag, but will almost never end up in the ‘S’ bag while a photon of 555 nanometres wavelength is equally likely to end up in the ‘M’ or ‘L’ bags. We determine colours according to the proportions of photons detected in each bag. It matters not at all what was the actual wavelength of the original photons, so long as they are sorted into the same proportions, we will see the same colour. However, the matter of the colour that we see is not straightforwardly a matter of measuring the proportion in S, M and L, because the human visual process is more complex than that. It should be noted that there is nothing fundamental about the use of three types of receptor. Birds usually have four, much more evenly spaced than human beings and therefore almost certainly have better colour vision than humans (see figure 2b). Very few mammals are tricromats – the evolutionary theory is that primates redeveloped trichromaticity by splitting the long wavelength cones, hence the very close response of the M and L cones.
Human colour processing
The understanding of colour vision is a major achievement of science. Imagine having to reverse engineer a digital camera on the basis only of what is told to you people viewing images that the camera produces. This is in effect what those striving to understand colour vision have had to do. The original proposal that colour vision was a tri-stimulus phenomenon was made by James Clerk Maxwell, the great Scottish physicist. Maxwell surmised, wrongly as it turned out, that any visible colour could be produced by mixing red, green and blue light, the so-called ‘primaries’. He produced the first permanent colour photograph in 1861, reproduced in figure 3, by making three separate exposures through red, green and blue filters and combining the result (a process which ultimately resulted in the Technicolor motion picture colour process).
Figure 3. The first colour photograph taken by James Clerk Maxwell
This was what would now be called the ‘RGB’ (for Red, Green, Blue) colour theory. However, the RGB theory failed to explain some of the oddities of human colour vision. For instance, if one can make any colour by mixing red, green and blue, why can one not achieve a greenish red? If one mixes green and red light the result is always seen as green or red or a completely different colour, yellow. Similarly, there is no bluish yellow. Yellow can be produced by mixing red and green light. Introducing blue light should produce a yellow that looks more blue, but instead it produces greys or white. The opponent theory suggests that the brain recognizes colour as two measures, one is a range of ‘redness’ or ‘greenness’, but not both. The second is a range of ‘blueness’ and ‘yellowness’, similarly mutually exclusive. Thus, colour is processed by the brain in only two ‘channels’ rather than three, the third channel being ‘brightness’. Figure 4 illustrates this state of affairs, and the way that the brightness or luminance is derived by adding together the three receptor channels.
Figure 4. Derviation of the three receptor channels
Colour reproduction
Photographers should be grateful that humans are not as visually blessed as the mantis shrimp, the eyes of which have 12 different types of colour receptor and can detect different polarisations as well. The sheer complexity of designing an accurate colour reproduction system for such visual capability may be one reason why the mantis shrimp has never developed photography. For humans, we just have to devise a way of separately stimulating our meagre three types of photo-receptor in such a way as to mimic the proportions resulting from light coming from the scene that we wish to reproduce. To do this, firstly we need to measure how the light emerging from the scene will stimulate the three types of receptor. The three stimuli measured do not have to be identical to the cone responses to be able to detect the same set of colours as the human eye, they have to satisfy what are called the ‘Luther-Ives’ conditions, after the two physicists who independently discovered them. These dictate that the response functions of the detectors must be linear combinations of the eye’s cone response functions. In practice, no real camera satisfies the Luther-Ives conditions, so no camera can detect exactly the same set of colours as can a human being. The opponent processes in the visual cortex provide the opportunity to make image processing more efficient.
Figure 5. The brain’s colour processing uses a luminance ‘channel’ and two ‘colour opponent’ channels.
For instance, typically we can detect more resolution in luminance than we can in the two colour ‘channels’, with the blue/yellow giving the least resolution. This is exploited in a number of ways. The file size of colour images may be reduced by storing luminance and two colour channels and reducing the resolution in the colour channels, as occurs in JPEG file formats and some others (such as Canon’s sRAW formats). Also, noise reduction may be applied differentially to colour and luminance channels, preserving detail in the luminance channel while smoothing noise in the colour channels. The Bayer matrix used in most cameras also exploits this characteristic by doubling the resolution of the luminance (green) channel with respect to the two colour (red and blue) channels.
