Image Color Picker

Pulling a palette out of a photograph is a harder problem than it sounds, because a photograph does not contain six colours — it contains tens of thousands. Reducing that to a handful means deciding which differences matter, and the obvious approaches decide badly.

Picking the most frequent exact colours fails because in a photo almost no two pixels are identical; you get six shades of the same sky. Rounding every channel to a fixed grid fails differently, producing colours that are near-misses of everything and matches of nothing. This uses median cut, which repeatedly splits the colour space along whichever axis is most spread out, so the palette adapts to the picture: a sunset spends its entries on oranges rather than on greens it does not contain.

Each colour is then reported with the share of the image it actually covers, and the swatches are sized to match. That ordering is what makes the first entry a genuine dominant colour — median cut produces its boxes in the order it happened to split them, which has nothing to do with prominence.

Transparent pixels are excluded. A logo on a transparent background is mostly alpha zero, and averaging those in builds a palette out of the parts nobody can see.

These are different things and the difference matters. The average colour of an image half red and half blue is purple — a colour that appears nowhere in it. Averaging is useful for a background tint that will not clash, and actively misleading as a description of what a picture looks like.

The dominant colour is the palette entry covering the most area, which for that same image is red or blue. That is almost always what people mean when they ask what colour something is, so it is what leads the palette here. The average is shown separately, and labelled.