While sharping a photograph, it really does not become any sharper because no algorithm in the world can create image data that has not been there before.
The algorithm detects edges, that is where areas of different color or brightness actually touch. In the areas surrounding these edges the contrast is enhanced.
Doing so, it well creates an impression of sharpness.
(Unsharp mask with various values for radius. The original is top left.)
The closer a pixel is to the edge, the stronger is the contrast enhancement. The further away it is, up to the radius given, the weaker the contrast enhancement will be. In the ideal case, this will really improve the sharpness, but often it does not.
When the radius does not match the radius of the actual blur then artifacts may visible. Such as Halos (A better word is brightness fringing, but most people just say halos.) as you see them on so many over-sharpened pictures.
This simplified example is sort of overdone just to put emphasis on the message behind this article. The original (top left) is perfectly sharp already but it may not appear quite sharp because of its rather low contrasts.
You need to understand that the sharpening radius must match the radius of the blurring that was there before. Unfortunately you do not know the blur radius. An additional challenge is that a photograph is typically not blurred with the same degree (radius) all over the picture. Typically the image is less sharp to the edges. Plus there is the depth of field effect. The farer away a sujet is from the exact focal pane, the blurrier it will appear. Therefore you will have to sharpen various areas of the image with different strength, because you want to sharpen strong enough for the sake of it on one hand and you want to avoid halos on the other.
The conclusion should be:
1) If an image is already sufficiently sharp, and please judge on it in some considerable distance to the full image, then don’t sharpen it at all.
2) If the image is blurred beyond rescue then sharpening would not help at all.
