In a prior blog entry I commented on how amazing image making is in light of technologies, materials, and manufacturing. I would like to continue along these lines and add something we never had in the old film/chemistry days.
Software.
To me this is as amazing as the physical implementation of sensors and cameras and lenses. In software we can implement just about any standard or process or idea we want.
There are algorithms for many aspects of image processing that directly effect how we view a photograph. Exposure, contrast, color profile, color depth, color management, image processing spaces (up to 32bit floating point!), tone mapping, film emulations, lens corrections, image alignment, perspective corrections, image stitching, high dynamic range, and... and... and... there are so many things implemented in software in various ways that it can be mind boggling.
It can be easy to be confused or frustrated and overwhelmed by all the options. What to they mean? What do they do? How do I take advantage of this selection compared to that one over there?
Considering one example from Rawtherapee, there are 19(!!!) options for demosaicing a RAW file on import into this software. Yes, certain selections can be made for users if you let the software automation retain control, but users are also free to select something else. Each and every option has a reason to exist. Understanding them can take time, certainly, but what a rich set of possibilities there are in just this one single step.
Looking at another example, there are at least 15(!!!) sharpen operators in the Open Source G'Mic. 15 different direct ways of sharpening an image. Each one is implemented based on a formula/algorithm that attempts to enhance some condition or another. If we add high by-pass filtering in a layer, there are even more image sharpening options, and I've not counted the sharpening operators in the Gimp (where I run G'Mic).
Taking this yet another step further, mixing various operators can refine an image processing sequence. In the case of the subject matter I currently enjoy photographing (automobiles and motorcycles) and taking into consideration digital enlargements (ie: going from 6000x4000pixels up to, say, 12,000x8000pixels) I've found the following sequence to give generally outstanding results.
- Open TIF RawTherapee output in the Gimp
- G'Mic DCCI2x upsize the image
- G'Mic Inverse Diffusion sharpen using just enough iterations to avoid introducing visible artifacts
- G'Mic Octave sharpen the image
In this case I use _two_ image sharpeners. One avoids pixelation (Inverse Diffusion) and the other gently sharpens subject edges.
Sometimes Octave is too strong. In which case I use one of two sharpen operations. One is applying a high by-pass filter layer blended using "Soft Light" over the base DCCI2x'd Inverse Diffusion sharpened image. This is brilliant for keeping pixelation to a minimum.
If that approach doesn't work and I need something stronger I use...
- Create a new G'Mic Gradient Norm layer from the base image
- Duplicate the base DCCI2x'd Inverse Diffusion sharpened image as another layer
- Copy the Gradient Norm layer into the mask of the duplicated base image
- Apply a hard G'Mic sharpener, such as Richardson Lucy
- Duplicate the sharpened layer to add further sharpness to the over all image
You have to see any and all of this to believe it. Which is another point I'd like to make. When using various sharpeners, I've found it really helps to watch the action of a sharpener at full pixel-peeping resolution.
Returning to the very start of the image processing pipeline, there is a sharpening operator in RawTherapee that I've found to be very useful. It's called "Capture Sharpen." It's best applied to low-ISO images where noise is at a minimum and is used to "correct" the effects of AA filters or image smearing due to slightly missed focus or slight camera movement.
The effect of Capture Sharpen can be dramatic and yields an image that I feel avoids any effects of being "over sharpened."
In the case of my comparison of a Sony FE 55mm f/1.8 with a c.1973 Nikon Nikkor-P (Xenar) 105mm f/2.5 pre-Ai, I thought I'd confirm these effects of Capture Sharpen. Here's what I found.
A comment about Capture Sharpen: This operator is used in certain RentWare, too, and sets the basis for their digital enlargement process in something I think they call "Smart Sharpen" and "Super Resolution". We can do the very same thing in Open Source Software. I described three ways to perform digital enlargements earlier in this post.
From this little comparison I see that Capture Sharpen applied to the Nikkor-P image brings the crispness/sharpness/contrast well into line with the native un-sharpened Sony FE 55mm. I can take an old lens and turn it into a digital-era Sharpness Monster. Applying Capture Sharpen to the Sony FE images takes images to another amazing level of sharpness.
How cool is all this? And this is just one little step in the image processing pipeline.
Software. I love it! Oh the possibilities...
No comments:
Post a Comment