I’m one of those outliers who use DxO instead of Adobe Lightroom and one of the reasons has been the way they automatically correct lens imperfections because they test each camera body and each lense. For instance, they say they correct vignetting because they know at every aperture and speed what is the darkness at corners and correct for it. I don’t know if that is true, but it seems to work pretty well.
But I’ve never really played with their lens softness parameters. I normally leave it on default. Sometimes when I have some motion blur, I will add some unsharp mask, but this isn’t too analytic. So given all the discussion about lenses and their resolution, I thought that I would learn more about exactly what is going on. There marketing says that they can recovery details and also microcontrast (that is sharpen the transition between dark and light. What they do is that they know where a lens is sharpest and have a “blur map” which takes that into account. They basically apply more sharpening at the corners than in the center.
Lense can also have astigmatism which means that the software is worst in a certain direction. That is softness might be worse at the lower left and better towars the upper right. And blur can be different for different colors so you want to apply different correction based on color. Finally darker regions have more noise, so you want to do less sharpening when it is darker. So given all this, how do you use DxO?
Here is their tutorial:
First some definitions:
- Sharpness. This is the ability to reproduce fine details and micro-contrasts.
- Focus. If you misfocus the lense itself.
Here are the steps
- Global. This is the amount of overall sharpness. It is set to -0.50. In the example, they move it to as much 0.8, so it is unclear what the normal range is
- Details. After you adjust the global, then you go to details and extracts fine detail (that is the microcontrast). In the example, they move it from 50 to 70 (no idea what the numbers mean). This is good for landscapes and architectural but on portraits, it will bring out small skin defacts.
- Bokeh. One problem with sharpenting is that it creates artifacts in the blurred backgrounds (called bokeh), so the Boken slige helps protect against that. So you need to look at the blurred background make sure there isn’t noise in the dark to light transitions and trun up Boken to make that disappear
In looking on the web for the correct values, there is quite a discussion about what appropriate values are:
- Fotomatti says that he uses 0 to 0.5 most of the time instead of -0.5 for global. Most of my photos just use the default, but I’ve been doing more pixel peeping at 1:1 and it does seem to help to go to at least 0. As Robgo2 says this is not USM, but is instead deconvolution and deblurring which is the same process as Photoshop Smart Sharpen, Focus Magic and InFocus but it takes into account lense specific distortions.
If you have a camera or lense not supported by DxO then you are back to using the unsharp mask features. Here is how to do it:
- Set the intensity to something like 200 to make everything generally have high contrast between light and dark. If you make it too high, you will get “halos” long the edges) and they leave the radius of sharpening and threshold to start sharpening at 4. These are standard USM parameters with a well known tutorial that basically says start at 100% and 1 pixel radius. The theory of USM is actually pretty interesting but basically, it works by blurring a image and applying it against it to find edges, then create overshoot around the edges to make things look sharper. 100% means that you do a 100% overshoot compared with the actual edge. Radius means how much you blur the original, so the smaller the radius, the smaller the size of the edge. And threshold means you set the darkest color you leave unenhanced. The negative is that you can see halos around edges if you apply too much, because the overshoot on the light side is so high it makes a bright white around the edge. The solution is either less amount or shrink the radius (so the halo is smalller)
- Sharpen the edges using edge offset, in the example, they set it to 200 and this will help even more along the edges
Now I also have Adobe and they have Adobe Camera Raw which also does lens correction in a preprocessing step before going to Photoshop for detailed editing or Lightroom to manage photos. Like DxO, they also distortion correction with a quick tutorial and you can also create custom profiels as well, but here are the values:
- Distortion. 100 means 100% of the correction. Seems like this makes sense
- Chromatic Aberration. 100% means all of the correction
- Vignetting means 100% of the profile
Also it looks like Adobe Camera Raw (ACR) does the same deconvolution/deblurring when you move the detail slider from 0 to 100 which moves from USM to deconvolution, but they do it based on lense characteristics.
The future however looks like it is in smart deblur. These are pretty sophisticated algorithms that basically assume that the image itself is in motion and tries to figure out the motion (eg the blur) and reverse it. Quite different that unsharp mask which is trying to make edges rink and be sharper. You can use the same idea to remove user shake because you can make some assumptions about how the image was shaken and then reverse it. You estimate or guess the blur kernel. This is the actual shake and then reverse it. The algorithm described in the last paper is actually pretty cool. Shows that you can estimate the motion of the camera pretty well by heuristics looking at edges. Dgrin.com has a pointer to smartdeblur.net which implements this as does focus magic (Windows only though!) including smartdeblur.net or Focus Magic which are probably good reasons to crank up my Windows VirtualBox or run wine on the Mac. As an aside I have a license to Focus Magic, but they are way behind on the Mac version, but are starting beta testing now.