Unblur an image: what the page does first
To unblur an image honestly you first have to know what kind of blur it is, because the kinds need different passes and the wrong one makes the picture worse. Before anything is offered, this page measures how definite the strongest edges are, how much noise sits in the flat areas, whether the frame carries a single direction of smear, and whether it is a two-tone page of text rather than a photograph. From those numbers it proposes one of six answers and opens the console on the matching pass.
| Diagnosis | What runs | The limit |
|---|---|---|
| Soft focus | Unsharp mask at a radius chosen from the measurement | Raises edge contrast; adds no detail |
| Camera shake | Wiener deconvolution against the estimated length and angle | Works when the movement was one straight, even path |
| Out of focus | A wider-radius sharpen with the halo clamps held in | Mild misfocus improves a little; heavy misfocus does not |
| Flat, not blurred | Clarity and structure (local contrast) | This is contrast, not blur |
| Text or document | Stroke-scale sharpening with paper texture left alone | Letters that merged in the photo stay merged |
| Too far gone | A conservative sharpen and a touch of clarity, labelled | Nothing recovers detail that was never recorded |
The suggestion is preselected, not imposed. Every card can be chosen, and every number the pass uses is visible and movable at the next station. The verdict line under the Found sentence says whether the chosen pass fits the measured blur, fits partly, or does not fit.
Why the verdict is sometimes no
The largest share of blurry photographs are out of focus or were taken in low light with a slow shutter and a moving subject. In both cases the fine detail was averaged together when the picture was taken. A page can steepen what is left, and this one will, but it will not print the word unblurred over a result that only looks crisper. When the measurement says the blur is beyond what can be inverted, the page says that first.
The camera-shake path is the one case with real recovery in it, and it is argued at length on the motion blur page, with the estimator's own accuracy published in the engine report. The written case for where deconvolution stops is in what deconvolution can and cannot recover.
Questions about unblurring a photo
Can any tool unblur a photo completely?
No. Blur is either a rearrangement of the detail or a loss of it. A straight, even camera shake rearranges it, and a deconvolution can partly reverse that. A lens focused on the wrong plane averages fine detail away, and what was averaged away is not in the file to bring back. Tools that appear to recover it are generating plausible detail. This page measures which case you have, says so, and runs the pass that suits it.
Why does it say my photo is too far gone?
Because the measurement says the blur is larger than any pass here can invert: a motion path longer than 40 pixels, or edges so faint that there is almost nothing left to sharpen. Rather than run a strong filter and call the result unblurred, the page offers a conservative sharpen and a little clarity, and labels it as exactly that. You can still pick any other card and judge the result yourself.
Is my photo uploaded?
No. The picture is decoded by your browser and every measurement and pass runs in this tab, in a worker on your own processor. There is no upload endpoint on this site to send it to. Disconnect from the network after the page loads and the whole job still works.
What is the difference between deblurring and sharpening?
Sharpening steepens the edges that are already there: it raises contrast across them and adds nothing. Deblurring, in the sense used here, inverts a known blur — it divides out the smear of an estimated motion path. The second can bring back some detail that the first cannot, but only when the blur fits the model. The saved file says which one ran: it ends in -deblurred when the motion solver was on and -sharpened when it was not.
Can it unblur a censored or pixelated image?
No, and nothing can. A deliberate mosaic or heavy blur placed over a face, a plate or a document replaces the detail with an average of it. The information is not hidden underneath; it was destroyed. This page does not attempt it.
