photo sharing and upload picture albums photo forums search pictures popular photos photography help login
Arn | profile | all galleries >> tests, samples, misc stuff >> Tutorial - sharpening tree view | thumbnails | slideshow

Tutorial - sharpening

Methods for sharpening

last updated 2006.03.28 (re: effect of LAB sharpening on colors)

Methods for sharpening


Curves and Levels ^
Increasing contrast and sharpening are basically the same thing. Sharpening is just contrast adjustment on the areas of image that have high contrast.
That's why contrast adjustment such as Levels and Curves adjustment of Photoshop increase the apparent amount of detail in the image.

In the following Curves example there is the typical S-curve that increases contrast. This particular S-curve is quite subtle, so experiment on different curves and see what kind of results you get. With Levels adjustment, you increase the contrast by sliding the slider on the slider towards right and (of course) the slider on the right towards left.
You can also try the "Auto" levels button or the Auto Color Correction options behind the "Options" button.

A simple sharpening example ^
The following image is a simple sharpening example, that demonstrates what basic sharpening looks like at pixel level. The image has been blown up 300% and the example is exaggerated. The upper half is sharpened and the bottom half is the original. The block of grey on the left has RGB values of 153/153/153 and the one on the right has 102/102/102. In the upper part, USM has increased contrast in the border area, where the two blocks of grey meet.

The standard way of sharpening an image, is using a sharpening filter like the Unsharp Mask (USM) in photoshop. People will usually tell you to use the USM with settings like radius: 0,3 amount: 100-300%, which is the appropriate range of values for a DSLR image with little in-camera or RAW sharpening. Treshold is a method of concentrating the sharpening on the high contrast areas of image and the value of threshold determines how big a difference in contrast between neighbouring pixels is required for sharpening to take place. Zero value sharpens the whole image. You might (for example) think of using threshold to minimize increase in noise caused by sharpening, but it is usually not a good idea, because use of threshold tends to produce a certain unevennes of sharpened areas / noise. The result will look unnatural. High pass sharpening (described later) for example produces better results, when wanting to avoid sharpening the whole image. I completely avoid the use of threshold.

USM with large radius ^
You can also try using Sharpening filter with large radius, by setting the values to something like radius: 50-100 and amount: 5-20%. That will work as general contrast adjustment.

High pass sharpening ^

Unlike the USM filter, you have to duplicate your image to two identical layers and apply the High pass filter to the duplicate layer. Then you change the blending mode of the duplicate layer to "hard light" or "linear light" and opacity to (for example) somewhere between 20-100%.

Below: Original image, high pass preview and final result after blending mode selection and opacity change.
The high pass sharpened example is a little oversharpened, but probably makes the example clearer.

Sharpening in LAB mode ^

When sharpening with USM, some slight color change (color fringing) can sometimes happen in the high contrast areas. This is because color data is affected by the sharpening. With LAB sharpening you can avoid this, because it will not sharpen the color channels, just the luminance data. This kind of sharpening may be useful when dealing with an image that already exhibits some fringing before sharpening. ...Although, there are better ways to deal directly with fringing.

As said before, sharpening (without masking, etc.) tends to inrease noise. Because LAB mode sharpening can be applied only to lightness channel (luminance data), leaving the color channels unsharpened, this method may result in slightly less color noise.

In practice the effect on noise is often quite slight, so that I have to blow an image up 300% and do some heavy sharpening before I can tell a significant difference. On other images the effect is more pronounced - certainly in those images that have loads of color noise to begin with. The effect on color fringing also depends on the image being sharpened.

Also, converting an 8bit RGB image to LAB and then converting back to RGB seems to result in the number of unique colors in the image being reduced. See Bruce Lindholm's page for a more thorough explanation here:
When I converted Bruces sample image to LAB and back (using relative colorimetric rendering), the result was a 69,4% loss of individual colors (5134605 colors remaining) which is for some reason significantly more colors remaining than in Bruce's example.

The effect is not as pronounced in everyday images, as it is in post processing done to the example image on Bruce's page - but could be significant anyway. You can test the effect by taking Bruce's image and converting it first to LAB and then back to RGB - then switch between unconverted and RGB-LAB-RGB converted image (you can make a layer for the converted image). You have to zoom in about 500% to best see the change.

I tried the RGB-LAB-RGB conversion on a typical RAW image that had the typical adjustments (like levels and curves) done in the RAW conversion program. The LAB conversion resulted in 23% loss of induvidual colors, but the effect was not very clear to the eye. I had to zoom 200-300% to the image and switch back and forth continuously between the original and the LAB converted image to see the difference at all. Without a reference image, the difference was impossible to see.

I also took the image and counted the amount of colours in various phases of sharpening:

In the last version, the original image was first converted to LAB, then sharpnened 400% 0,3 in lightness channel and then converted back to RGB.

The increase in the amount of colours after sharpening is quite logical if you just take a look at the simple sharpening example and see what sharpening does to pixels on the contrast border. USM actually generates "more colours" in to the image by altering the brightness of the pixels where sharpening is applied. :)

Fade tool ^
Fade tool can be used to do the same as LAB sharpening. After sharpening with USM or Photoshop's Smart Sharpen, select from the drop down menu "Edit/Fade tool" and Mode: Luminosity with opacity of 100%. This will result in the sharpening being aplied to luminance values only. Pretty much what LAB mode sharpening does and a lot simpler. On Focalblade sharpening I can't make any practical difference with the Fade tool, at least if I don't sharpen with outrageous values like amount:800% and radius:5 pixels. Looks like Focalblade works on luminance data like LAB sharpening and doesn't produce color fringing or color noise in a way that USM does. So it's useless to apply Fade tool to reduce noise after Focalblade sharpening.

Different cameras and in-camera sharpening ^
As I said before, these values of radius and amount apply for a typical DSLR image with minimal sharpening, such as the Canon 20D / 5D / 350D, Nikon D200, D70, etc.

If you use internal sharpening of a DSLR camera (in jpgs) or sharpen the image in a RAW converter, the values might have to be lower. For best results I recommend using minimal in-camera sharpening, not using the sharpening methods of the RAW converter and always sharpen in post processing. But your mileage may vary and you may wish disregard this advice for the sake of easing and speeding your workflow.

If you use a typical digicam, the values should DEFINITELY be lower, because the typical digicams apply far more sharpening to an image than a typical DSLR does to a JPG file. You may be able to produce an unsharpened file from a digicam if it has the option of saving the file in RAW.

*Let it be said, that sharpening does not ADD any detail to the image. It just changes the image, so that it appears to have more detail to the eye.

Monitor calibration ^
Remember that your monitor and it's calibration/settings affect the apparent sharpness of the image, just as they affect apparent contrast and color balance.
Your monitor can lie.

Side effects of sharpening ^
USM tool (Unsharp mask) of your post processing program is a good tool, but that may result in some side effects - mainly halos and increased noise. Halos of black or white form around the high contrast areas, basically because too much sharpening is applied there. The noise is increased, because the Unsharp mask does not limit the contrast adjustment to the high contrast areas of the image. An equal amount of sharpening is applied everywhere in the image and this also increases the contrast of digital noise.

Other forms of general contrast adjustment like Levels and Curves may also increase the appearance of noise, for the afore mentioned reasons.

an image with black halos and increased noise, due to oversharpening:

Neatimage ^
A bit more sophisticated way of sharpening can be found from the noise removal program Neatimage

The main use for the program is the removal of excessive noise, but it can be of some use in sharpening.
It's built-in sharpening methods include masking the the high contrast areas, so that the sharpening can be applied where you need it most. This prevents increased noise elsewhere in the image.

masking looks something like this:

Here's a direct link to the program's user guide and it's sharpening section:

I use 'Conservative' settings in NI and mostly the 'High' slider.

Focalblade ^
One of the best sharpening tools out there is the Focalblade plugin for Photoshop:

The main benefits of Focalblade are:
- You can set the amount of sharpening separately for the whole image and high contrast areas. This results in less noise on flat areas of image as when using Neatimage.
- You can reduce the amount of haloes in the programs 'Fix' settings. Halo artifacts are usually the thing that sets the limit of the amount of sharpening used.

RAW conversion and sharpening ^
About RAW converters and sharpening: I am yet to see a RAW converter with good enough sharpening tool, as to be actually useful for sharpening an image in post processing. Better results can be achieved with other methods. At least in my workflow.

The following image demonstrates the typical "blockiness" that the sharpening of RawShooter causes. The image is oversharpened for sake of clarity. Let it be said that RawShooter Premium is in my opinion probably the best RAW converter available at the moment and worth every penny. And I have tried just about every one of them. Other RAW conversion programs seem to produce similar artifacts when sharpening, but I can't say for sure how common this trouble is among RAW converters as I haven't found the time or the interest to test this particular phenomenon among different converters. It could be that the effect comes from RawShooter trying to maximize the amount of detail that the sharpening produces and lack of antialiasing. Or something. I'm not going to go guessing, it's enough that for me the sharpening result is not very natural.

How much to sharpen? ^
The following image just demonstrates two levels of sharpening, the upper image has more sharpening applied. The amount that you feel is appropriate for a particular image depends wholly on your taste, what you are trying to achieve and the type of the image. The upper image might be considered to have just the right amount of sharpening, so that disturbing halos or noise haven't been introduced yet. The lower image might be considered a tad soft (not enough sharpening). But then again, if you want your cat to look soft and cuddly, it might be that the lower image has just the right amount of - or even too much - sharpening! Purely a matter of opinion and what kind of look you are going for.

and the following thumbnail links to the same image as above,
it is left there purely because otherwise this tutorial won't have a thumbnail shown in the parent directory...

Snickers sharpening comparison strong vs average amount
Snickers sharpening comparison strong vs average amount