Noise & edit-region maps
Find edited or spliced regions in a photo
Drop in a photo and switch between three classic forensic views. Each one throws away the ordinary picture and shows a different residual, so a region that was pasted, painted over or cloned tends to stand out from its surroundings. Read them as leads to inspect by eye, not proof.
Runs entirely in your browser. Your image is never uploaded or stored.
What each view shows
| View | What it does | What to look for |
|---|---|---|
| Noise | Subtracts a blurred copy to leave only the fine grain | A region whose grain is smoother, busier or differently textured than the rest |
| Luminance gradient | Colours each pixel by the direction its brightness is changing | Edges lit from a different direction, or a patch whose gradients do not match the scene |
| Level sweep | Isolates a single brightness band and stretches it to full contrast | Hidden structure at one brightness, or a pasted area that glows at a level the rest does not |
How to read them
None of these views is a verdict. They amplify what a photo hides from the eye, so that an area with a different history has a chance to show itself. In the noise view an untouched photo has a fairly even grain across the whole frame, and a region that is unusually smooth or unusually busy is worth a second look, though soft focus, sky and heavy compression all flatten noise innocently. The luminance-gradient view turns the direction of every small brightness change into a colour, so a consistently lit scene shows smooth colour transitions and a spliced object often carries edges coloured differently from its surroundings. The level sweep walks through the brightness range one band at a time, which can surface faint structure, banding or a patch that was filled at a single value.
Where they fail
Every one of these maps lights up for innocent reasons. High-contrast edges, text and saturated colour always look busy; a smooth sky always looks clean; and repeated resaving through social media or a chat app scrubs the fine noise these views depend on. A striking pattern here is a reason to corroborate with another method, never a conclusion on its own. Pair this with the photo edit check, the JPEG Ghost tool and the metadata viewer before drawing any conclusion.
Sources
- Krawetz, N. (2007). A Picture's Worth: Digital Image Analysis and Forensics. Black Hat USA 2007.