Three numbers per pixel, and why green counts most

Every colour image is three grey images

Open the Channels panel in Photoshop, Photopea or GIMP and switch off two of them. What is left is a greyscale picture. That is the whole structure of a colour file: three separate brightness maps, one for red, one for green, one for blue, stacked and displayed together.

An 8-bit file gives each channel 256 levels, 0 to 255. Three channels of 256 levels is 16,777,216 possible combinations, which is where the phrase “16.7 million colours” comes from. A hex code is just those three numbers in base 16: #FF0000 is red 255, green 0, blue 0. #808080 is 128 in all three, which is why equal values always give you a neutral grey.

This is additive colour. It describes light being emitted, so all three at maximum gives white and all three at zero gives black. Ink works the other way round (it subtracts light from white paper), which is why print uses CMYK and why some screen colours cannot be printed at all. That is later in this module.

The channels are not equal, and your eye is why

Here is the fact that changes how you work. Brightness is not the average of the three numbers. For sRGB images (on linearised values) and HD video (ITU-R BT.709), the weighting is:

luminance = 0.2126 × R + 0.7152 × G + 0.0722 × B

By these weights green carries roughly 71 per cent of the perceived brightness, red 21 per cent, blue 7 per cent. JPEG files and standard-definition video use the older BT.601 weights of 0.299, 0.587 and 0.114. Both say the same thing: green counts most, blue least. This is not a convention somebody picked. The human retina has far more cones sensitive to the middle of the spectrum than to the ends, so a pure green patch looks much brighter to you than a pure blue patch of the same numeric value.

Two practical consequences follow immediately.

The green channel is usually your cleanest one. Camera sensors have twice as many green photosites as red or blue, so the green channel has the least noise and the most real detail. When you need a channel to build a mask from, start by looking at green. The blue channel is usually the noisiest, which is also why blue skies band first.

Desaturating by averaging is wrong. Set a Hue/Saturation layer to −100 and some tools average the channels. A red jumper and a blue jumper that look obviously different in colour come out as near-identical greys, and the picture goes flat. Use a proper Black and White adjustment, which lets you set the weight per channel. There is a whole lesson on this later.

HSB is a steering wheel, not a storage format

The colour picker offers you Hue, Saturation and Brightness. Nothing in the file is stored that way. HSB is a re-parameterisation of the same three numbers into controls that match how people talk: hue as an angle round a wheel, saturation as distance from grey, brightness as overall level.

It is easier to work with (“make it slightly less orange” is one slider in HSB and three in RGB) but it hides a trap. HSB brightness is not the luminance formula above. Pure yellow at HSB brightness 100 and pure blue at HSB brightness 100 differ by a factor of almost thirteen in actual light (0.93 against 0.07 by the formula above). Move a hue slider and the perceived lightness moves with it, which is why a hue shift on a logo often needs a lightness correction afterwards.

Look at your channels once and you will keep doing it

A two-minute habit worth building. In any editor, cycle the channels on a portrait:

• Red: skin goes pale and smooth, because skin reflects a lot of red. Blemishes almost vanish. This is why frequency separation and skin work often lean on red.
• Green: the most “normal” looking, and the sharpest.
• Blue: skin goes blotchy and dark, hair separates hard from a pale background, noise is obvious.

That blue channel, ugly as it is, is often the fastest route to a hair mask. In GIMP: Colours > Components > Decompose to see all three as separate layers. Photopea and Photoshop both have the Channels panel with the same keyboard shortcuts.

The honest limit: three numbers are a model of you, not of light

Real light arriving at a surface is a continuous spectrum: energy at every wavelength. Three numbers cannot record a spectrum. They record how strongly that spectrum would excite three particular kinds of receptor.

The consequence has a name: metamerism. Two objects with completely different spectra can produce identical RGB values under one light and visibly different ones under another. This is the shirt and trousers that match under shop fluorescents and clash in daylight, and it is why a printed swatch can match a screen under one lamp and not another.

No amount of editing fixes it, because the information was never in the file. It is also why print work is checked under a standardised viewing light rather than whatever bulb is in the room, and why a client who approves a colour on a phone in a café has not really approved anything.