Median or edge-preserving — which should I use?

Use the median for discrete specks: dust on a scan, hot pixels, salt-and-pepper noise, since a single stray value is never the middle of its neighbourhood and disappears outright. Use edge-preserving for even grain across flat areas, since it averages only neighbours already close in colour, so grain inside that window averages away while the far side of an edge is never counted.

What can denoising not do?

Neither filter can put back the detail the noise was sitting on top of. From a radius of 3 upward both start eating small text, fur and film grain along with the noise, which is exactly why the preview runs at one image pixel per screen pixel rather than as a fitted thumbnail.

Is it slow on large photographs?

It can be. Both filters read a window of source pixels around every pixel they write, and a 7-by-7 window over an 8-megapixel photo takes a few seconds a file. The median uses a sliding histogram, so its cost grows with the radius rather than with the square of it. The export always runs at full resolution however small the preview window is.

Denoise

Denoise removes speckle and sensor noise in your browser using two filters: a median, which replaces each pixel with the middle value of its neighbours, and an edge-preserving average, which only averages neighbours whose colour is already close to the centre pixel. A median removes a stuck pixel completely — a single speck is never the middle of anything — while a straight edge survives, because it is the majority on both sides of itself. The preview runs at 1:1, since a shrunken preview averages the speckle away and makes every setting look clean.

How it works

Drop your images Drag one or more images in. They are decoded and the preview opens at full size.
Pick a method Median removes specks, dust and stuck pixels. Edge-preserving smooths grain across flat areas by averaging only neighbours of a similar colour.
Set the radius Radius 1 is a 3-by-3 window and it goes up to 5, an 11-by-11 window. Every step up removes larger clumps and takes more fine texture with it.
Set the colour threshold This appears for the edge-preserving filter only: how different a neighbour may be, in levels out of 255, and still be averaged in. Below it is noise, above it is an edge. The default is 28; at zero almost nothing changes.
Check at 1:1, then apply Drag or arrow-key around the preview before running the batch, then apply and take one file or a ZIP.

Questions

Median or edge-preserving — which should I use?: Use the median for discrete specks: dust on a scan, hot pixels, salt-and-pepper noise, since a single stray value is never the middle of its neighbourhood and disappears outright. Use edge-preserving for even grain across flat areas, since it averages only neighbours already close in colour, so grain inside that window averages away while the far side of an edge is never counted.
What can denoising not do?: Neither filter can put back the detail the noise was sitting on top of. From a radius of 3 upward both start eating small text, fur and film grain along with the noise, which is exactly why the preview runs at one image pixel per screen pixel rather than as a fitted thumbnail.
Is it slow on large photographs?: It can be. Both filters read a window of source pixels around every pixel they write, and a 7-by-7 window over an 8-megapixel photo takes a few seconds a file. The median uses a sliding histogram, so its cost grows with the radius rather than with the square of it. The export always runs at full resolution however small the preview window is.

Denoise

How it works

Questions

Every tool