How Does a Screen Make Colour? Three Lights and an Eye That Can Be Fooled

The retina contains three types of cone cell, each most sensitive to a different part of the spectrum, roughly in the long, medium and short wavelength ranges, and the brain computes colour from the ratio of their three responses. Because the output is only three numbers, any light that produces the same three numbers looks identical, whatever its actual spectral composition, and two such lights are called metamers. Pure yellow light stimulates the long and medium cones in a particular ratio; a mixture of red and green light can produce exactly the same ratio, and the eye has no mechanism to tell them apart. This is why three primaries suffice for a display and why the same trick does not work for a machine that measures spectra. It also means a display can never reproduce every colour a human can see, since the set of colours reachable by mixing three fixed primaries is a triangle inside the curved boundary of human vision, and a wider triangle needs more saturated primaries.

The competing panel types differ in where the light comes from and how it is controlled:

• •Liquid crystal displays have a constant white backlight, and each subpixel contains liquid crystal that twists to rotate the polarisation of light passing through it, which either lets it through a second polarising filter or blocks it. A coloured filter over each subpixel selects red, green or blue
• •Because the backlight never fully switches off, the darkest a liquid crystal panel can get is a dark grey, which is why contrast is its main weakness, partly addressed by dividing the backlight into dimming zones
• •Organic light-emitting diode panels have no backlight, since each subpixel generates its own light and can be switched off entirely, giving true black and very high contrast at the cost of a risk of permanent image retention where static elements sit for long periods
• •Quantum dot layers use nanocrystals whose size determines the exact wavelength they emit, converting blue backlight into extremely pure red and green, which widens the reachable colour range
• •Micro-LED, the newest approach, uses microscopic inorganic light-emitting diodes per subpixel, combining self-emission with high brightness and long life, and is currently very expensive to manufacture
• •Electronic paper works differently again, moving pigment particles in tiny capsules and reflecting ambient light rather than emitting any, which is why it is readable in sunlight and needs no power to hold an image

A number in an image file means nothing until a colour space says what it refers to. The standard for decades has been sRGB, defined in 1996 to match the behaviour of the cathode ray tube monitors of the time, and it covers a relatively small portion of visible colour. Wider spaces followed, including Adobe RGB for printing, DCI-P3 for digital cinema and now for most phones and laptops, and Rec.2020 for ultra-high-definition broadcast, which is wider than any current display can fully show. Colour management is the system of profiles that translates between a file's space and a display's actual capability, and its absence is why the same image can look different in two applications on the same screen. A further complication is the transfer function, the relationship between the stored number and the emitted light, which is deliberately non-linear because human perception of brightness is non-linear, so that the available bits are spent where the eye can see differences.

The more recent change is to brightness rather than colour. Traditional displays were designed around a reference brightness of about a hundred candelas per square metre, which is roughly a sheet of white paper in an office, and everything brighter in a scene, a sunlit window, a specular reflection, a flame, was compressed into that range. High dynamic range raises the ceiling to a thousand or several thousand, encodes the signal with a transfer function designed around the limits of human perception rather than around a display technology, and carries metadata describing how a particular scene should be mapped onto whatever the actual screen can do. The practical result is that highlights look like light sources rather than like pale grey, which is a more noticeable improvement than additional resolution. It also exposes the weakness of each panel type clearly, since a liquid crystal display producing a thousand candelas in a small bright area will leak light around it, and an organic panel will hold the black perfectly and struggle to sustain peak brightness across a large area.

Almost every design decision in the field is a calculation about perception. Resolution beyond roughly sixty pixels per degree of visual angle is invisible at normal viewing distance, which is why the useful resolution of a screen depends entirely on how close you sit and why phone and television specifications are not comparable. Colour subsampling exploits the fact that the eye resolves brightness detail far better than colour detail, so video compression stores colour at half or a quarter of the resolution of brightness and nobody notices, which is a large share of why video files are as small as they are. Frame rate requirements come from the perception of motion rather than from any property of the image. And the three-primary system itself is the deepest example, since it works only because human colour vision throws away almost all the information in the spectrum before the brain ever sees it.

Screens produce colour by mixing three primaries because the eye reports colour as the ratio of responses from three cone types, so different spectra that produce the same ratio are indistinguishable. Liquid crystal panels block a constant backlight and cannot reach true black, organic panels emit per pixel and can, and quantum dots purify the primaries to widen the reachable range. A colour space defines what a stored number means, high dynamic range raises the brightness ceiling, and compression exploits the eye's poor resolution for colour detail.