Sharpen Image
Choose an image
Pick a file above and the sharpened version appears here. Nothing is uploaded. The image is read and processed in this page.
Sharpening works by amplifying the difference between each pixel and its neighbours. The page shows the actual 3×3 matrix being applied as you move the slider, because sharpening is one of those operations that makes immediate sense once you can see the numbers.
How it works
Convolution slides a small matrix over the image. Each output pixel is the weighted sum of the pixel underneath and its neighbours. A sharpening kernel gives the centre a large positive weight and its neighbours negative ones, so wherever a pixel differs from its surroundings that difference is exaggerated.
centre = 1 + 4a, each of the four neighbours = -a
- a
- the amount, from 0 for no change upwards
- sum
- the weights always total exactly 1, which keeps overall brightness unchanged
- In a flat area every neighbour equals the centre, the negatives cancel the positive, and the pixel comes out unchanged. Sharpening only affects edges.
- At an edge the difference is amplified, making the light side lighter and the dark side darker.
- Push it too far and that becomes a visible halo, a bright outline on one side of every edge, which is the signature of over-sharpening.
- Edge pixels clamp to the nearest in-bounds neighbour, so the image does not gain a dark border.
Sharpening cannot recover detail that is not there. A blurred photograph has lost information, and amplifying local contrast makes the remaining edges more prominent without restoring anything. It works well on slight softness from resizing and badly on genuine motion blur.
Examples
The default amount
Amount
60%
Result
Centre 3.40, neighbours -0.60
1 + 4(0.6) = 3.4, and the four neighbours are -0.6 each. They sum to 1, so brightness is preserved.
No sharpening
Amount
0%
Result
Centre 1.00, neighbours 0.00
The identity kernel. Every pixel is copied unchanged, which is a useful check that the maths is right.
Over-sharpened
Amount
180%
Result
Centre 8.20, neighbours -1.80
Visible halos along every edge and amplified noise in flat areas. Still brightness-preserving, just ugly.
Frequently asked questions
Can this fix a blurry photo?
No. Blur destroys information, and sharpening only amplifies local contrast in what remains. It works well on the slight softness left by resizing or by a lens, and does nothing useful for motion blur or a badly missed focus. Those need the detail that was never recorded.
What are the white outlines around edges?
Halos, and they mean the amount is too high. The kernel pushes the light side of every edge lighter and the dark side darker; past a point that overshoot becomes a visible outline. Back off until they disappear, usually somewhere below 100%.
Why does the image get noisier as I sharpen?
Because sensor noise is pixel-to-pixel variation, which is exactly what a sharpening kernel amplifies. The effect is worst in flat areas such as skies. Reduce noise first if you have a tool for it, then sharpen.
Why do the kernel weights add up to 1?
So that overall brightness is unchanged. If they summed to more than 1 the image would get brighter, and less than 1 darker. Keeping the total at exactly 1 means the operation redistributes contrast without shifting exposure.
Should I sharpen before or after resizing?
After. Downscaling softens an image, so sharpening first means the softening undoes your work and the remaining artefacts get scaled with it. Resize to final dimensions, then sharpen lightly.