Colour is a perceptual experience, not a directly measurable physical phenomena. Of course there are underlying physical phenomena creating our experience, but we can experience the same physical phenomenon differently under different circumstances. Similarly we may experience two different physical phenomena as appearing the same. In general, there is no one-to-one mapping between the spectrum of the light reflected into our eyes and the colour as we perceive it.

A fundamental assumption is that humans exhibit colour vision to give us information about the surface properties of objects. The stability and reliability of colour information is important to our understanding of the world. The problem of colour stability arises because the light reaching our eyes from an object is the product of the object's surface reflectance and the spectrum of the light illuminating the object.

We do not have direct access to the properties of the incident light, so somehow we must estimate them from the light we observe from the object. To make matters worse, our eyes only measure the spectrum at extremely low resolution. Light energy is a continuous phenomenon that the human visual system samples over the wavelength range of approximately 300 - 700 nanometres using three sensors. Thus, the eye's output signal is a discrete triplet of colour values.

All commonly used imaging systems, from digital cameras, to scanners and displays, use a similar 3-band colour encoding scheme since their purpose is to provide input to the human visual system. One consequence of this colour encoding scheme is that there are many spectra that can lead to the same triplet of responses. Such spectra are called metamers, and reflect the fact that the relationship between spectra and sensor triplets is many-to-one. We cannot perfectly reconstruct the original spectrum from its reduced counterpart, because information has been lost that cannot be recovered. There is no hope of recovering the precise surface reflectance properties of surfaces in a scene from a 3-band image of that scene. Nor is there any hope of recovering the exact spectrum of the light illuminating the scene. The best that we can expect is a 3-parameter specification of the reflectance and illumination properties.

Another problem is that a change in the illumination can lead to a very different sensor triplet response, even though the surface properties of the illuminated object remain the same. This makes identification of objects using just sensor triplet information unreliable.