What is image compression?

An uncompressed photograph is a straightforward list: for every pixel, how much red, green and blue it contains. A 12-megapixel photo is twelve million pixels at three bytes each, which is about 36 MB before anything clever happens.

Almost none of that data is surprising. The sky in the top-left corner is very close to the sky next to it. A wall is thousands of nearly identical pixels in a row. Compression exists because a description of an image is almost always shorter than a list of its pixels — and because a great deal of what is in that list is below the threshold at which human vision notices anything.

Every compression format is a different answer to two questions: which patterns are worth describing rather than listing, and how much of the detail can be dropped before someone complains.

Lossless compression finds redundancy and encodes it more efficiently. If a row contains four hundred identical pixels, it records "this colour, four hundred times" instead of four hundred copies. Decode the file and you get back precisely what went in, byte for byte.

PNG works this way, and so does a ZIP archive. The guarantee is absolute, which is why it is the right choice for screenshots, diagrams, logos and anything you will edit again.

The limitation is equally absolute. An algorithm forbidden from discarding anything can only exploit redundancy that is genuinely present. A photograph, full of sensor noise where almost every pixel differs slightly from its neighbour, offers very little — which is why photographs saved as PNG are enormous.

Lossy compression throws information away on purpose, choosing what to lose based on how human vision works.

Two facts drive nearly all of it. First, we notice changes in brightness far more than changes in colour, so colour can be stored at lower resolution. Second, we notice broad shapes far more than fine texture, so fine texture can be described crudely or dropped entirely.

JPEG, lossy WebP and AVIF all exploit both. The result is a file five to ten times smaller than a lossless one, and — at sensible settings — an image you cannot tell apart from the original at the size it is actually displayed.

The cost is that the discarded detail is gone. There is no setting that recovers it, and no tool that can reconstruct it later.

Before any encoder setting, there is resolution. File size scales roughly with pixel count, so halving both the width and the height leaves a quarter of the pixels and close to a quarter of the file.

A 4000-pixel-wide photograph displayed in an 800-pixel column is carrying twenty-five times the data it can possibly show. No quality slider recovers that waste — only resizing does. Reduce the dimensions to what will actually be displayed first, and then think about quality.

It cannot add detail. Enlarging a small image produces a soft large image, because the information was never recorded.

It cannot reverse itself. A JPEG re-saved as PNG is a lossless copy of a lossy image, not a restored original.

And it cannot fix a file that was the wrong format to begin with. A screenshot saved as JPEG has haloes around every letter, and compressing it harder only deepens them. The fix is to change format, not settings.