The effects of this tool are only visible at a preview scale of 1:1 or more. Use a detail window (click on the icon under the main preview panel) to inspect a part of the image, or zoom the main preview to 100% (also called 1:1) .
This article describes the tool called "Sharpening", however RawTherapee contains other tools which can be used to perform various types of sharpening - see Edges and Microcontrast and the Wavelets tools.
The Sharpening tool is applied to the full image, before the Resize tool. If you would like to apply sharpening after resizing, use the Post-Resize Sharpening tool which you will find inside the Resize tool.
1 Unsharp Mask
Unsharp masking (USM) is a technique used to increase the apparent acutance (edge contrast) of an image, making it appear clearer, even though it technically does not really sharpen the image. It makes use of several phenomena of the human visual system in order to accomplish this effect, such as the Cornsweet illusion and Mach bands. Though unsharp masking in other software is easily prone to causing halos, RawTherapee has a unique threshold slider which allows you to achieve a superb sharpening effect with a minimal risk of halos.
The Radius determines the size of the details being amplified and, consequently, relates to the width of the sharpening halo. In general the quality of sharpening is best if the sharpening radius is smaller. For low ISO images that are in focus and without motion blur a value of 0.5-0.7 is satisfactory.
The Amount parameter controls the strength of the sharpening.
The Threshold tool helps to suppress noise amplification and haloing and to confine sharpening to a desired tonal range. The Threshold tool allows one to create a curve via which the sharpening is applied. The vertical axis corresponds to opacity: 0% at the bottom (transparent, sharpening not visible), 100% at the top (opaque, sharpening visible). The horizontal axis corresponds to luminosity: select the tonal range that will get sharpened - the darkest tones are on the left, progressing to white tones on the right. As mentioned in the tooltip, to move each of the points in the threshold tool individually, hold the Shift key before clicking on a point with your mouse. Holding the Ctrl key while moving a point with the mouse allows for very fine movements.
When moving the right pair of sliders to the left side, sharpening is reduced in the highlights. When moving the left pair of sliders to the right side, sharpening is reduced in the shadows and minimizes amplification of dark noise.
The default threshold values will protect from over-sharpening and haloing in most cases and limit the sharpening effect to mid-tones. In the example screenshot, the blackest tones have no USM applied, then USM is applied to a broad range of tones from dark to light, and the strength of USM gradually drops off from maximal at the mid-tones to none at the whitest tones, so as to prevent noise amplification and haloing.
1.4 Sharpen Only Edges
If you activate "Sharpen only edges" then uniform areas will not be sharpened. This is useful when sharpening noisy photos.
Two new sliders appear as well:
The Radius is used for noise detection. If the noise is low, a lower radius can be used, and vice-versa. A higher radius slows down the image processing.
1.4.2 Edge Tolerance
Edge Tolerance determines how much a pixel has to differ from its neighbor to be considered as an edge and not as noise. It is very similar to the USM Threshold parameter and has a high impact on the visual quality. For low ISO (low noise) images use 1000 or less, for high ISO images use 2500-3000 or even more.
1.5 Halo Control
Halo Control is used to avoid halo effects around light objects when sharpening too aggressively. When activated, a new slider appears:
At 100 it works at maximum, reducing the visual impact of the USM filter.
2 RL Deconvolution
RL deconvolution is named after the makers of this algorithm, Richardson and Lucy. It uses the point spread function (PSF) to deconvolve (reverse) the effects of Gaussian-like blur. In reality, the blur produced by the lens and by motion may differ from Gaussian blur significantly, therefore some artifacts, such as halos, may appear when the radius diverges too far from the type of blur in the actual image, and when then effect is too strong.
Controls the blend factor between the unsharpened image and the sharpened one.
Damping reduces the effect of the deconvolution at each iteration. It has the effect of preventing sharpening of the finest details. Use it if the sharpened image has too much "bite" at the finest level.
RL Deconvolution is an iterative algorithm; it requires being repeated in order to achieve the intended results. Each repetition of the process is called an "iteration", and the result of one iteration is used as the starting point of the next iteration. While each iteration removes blur, it also increases processing time and the likelihood that halo artifacts will appear, so you need to find the perfect balance through trial and error - the default value should be fine for most cases.