Thursday, August 11, 2022

Another Up-Sizing look ~ automobiles in the wild and out on the streets of Paris

Years ago (as in 14 at this point) The Online Photographer had an article about how a photographer took a 4 megapixel sensor'd image and made glorious 13x19 inch prints.

 Here's what was said about the image - "This rider was going about 75 mph when the photo was taken, and you can see every stitch, vent perforation, and the pebble texture of the leather with excellent detail and clarity."

This has stuck with me over the years.  I've written several times about trying to simulate a similar up-sizing process using the Gimp.  My first attempt involved UnSharp Mask sharpening the basic file, up-sizing using the Cubic operation, and then USM sharpening again.  It was an interesting process, but there was always something slightly soft about my final results.

Two years ago I saw that RawTherapee came out with a "Capture Sharpen" function.  I didn't think about it too much at first about how it might related to up-sizing an image.  All I knew was that my images took on a sudden and happy increase in "sharpness" in their native sizes.

I guess I could step back and say that my early Canon DSLR images, at the time, felt "sharp" enough to me.  Then I experienced Sony's APS-C sensors and realized I my Canon gear was lacking.  More recently I've added a couple Sony full frame sensor'd A7 cameras to the Box of Toys.  These take to "Capture Sharpen" like ducks to water and are far sharper than any of my old Canon files that are also "Capture Sharpened".  If there is anything sharper, please, please show it to me.  I'd like to see it.

In 10 years I've moved from Canon "acceptable", "ya, that's pretty OK" sharpness to Sony "oh my ever loving gawd!" levels of sharpness.

When is "enough" enough?  I've been wrangling over the added cost of moving up the Food Chain a bit to acquire a 42mpixel Sony A7R2.  The additional pixels could make life better, right?  More, better, happiness.  Stay with the "in crowd" to maximize flexibility and overall image quality.  Perhaps rather shallow justifications for buying more camera equipment.

In similar time, or should I say "just in time", I stumbled across an interesting video where a guy makes rather large pleasing prints from an old 10mpixel Leica M8.  It immediately reminded me that perhaps I hadn't fully explored careful up-sizing processes.  

Between Stephen Sharf's process notes and seeing this video I then felt I might even save a few Louis d'Or or Pistoles by staying with a lower cost but still (hopefully) viable 24mpixel sensor'd solution.  For the price of one used Sony A7R2 (around 1100USD) I could have three used Sony A7 (around 400USD each).

Borrowing from 14 years ago to consider the idea of making 42mpixel sized prints by careful up-sizing 24mpixel image might prove "interesting."  Stephen Sharf  started with a 4mpixel (2464 x 1648 pixels) Canon 1D image and ended up with a 6840pixel x 4680pixel print file that looked, by all accounts, to be wonderful at 13x19 inches.

"...

1. Each image is sharpened upon import into Photoshop using the Photokit Sharpener "Capture Sharpen" macro to recover detail lost by the sensor (effectively infinite number of photons, finite number of pixels).

2. Each image is then upressed using Bicubic Smoother in PS to give the pixel dimensions at 360 ppi.

3. The image is then sharpened for printing using Photokit Sharpener using the Inkjet, 360 dpi, Glossy sharpening macro.

4. The image is then exported to ImagePrint, a RIP, and printed on Stephen's old warhorse Epson 2400 using the appropriate color profile and ImagePrint to drive the printer.

5. The photo is then printed on InkJet Art Microceramic Lustre..."

Looking at my up-sizing process I thought about improvements I might make.  For the first sharpening step I could use RawTherapee's "Capture Sharpen" in place of the Gimp's various "smart sharpen" operators.

Second, I could use the Gimp's "NoHalo" up-sizing operator.  This would replace the "Cubic" operator that I had been using, and recently found to be soft compared with "NoHalo".  

Third, I could try various Gimp G'Mic sharpening operators to see if there was something demonstrably better than "UnSharp Mask."  To this end I find I like the G'Mic "Inverse Diffusion" sharpening operator.  I think it's really nice, particularly if I put a sharpened copy of the image in a layer and lower the opacity while observing the effect at 100 to 200 percent viewing sizes.

To test all this I took an image from a recent la traversee de Paris that I'd made using flash fill (three flash units, in fact, to try and keep up with the sun), a Sony A7, and a pretty little Nikon Nikkor 24mm f/2.8 Ai lens.

 

Renault Alpine ~ la traversee de Paris estivale ~ 2022

 

Here is what I found.

Rawtherapee Capture Sharen Comparison

 

Keep in mind, I'm well aware of the fact I'm not adding ANY information to an up-sized file.  All I'm doing is smoothing ("spackeling", if you prefer) the transitions between expanded pixels.  This is, afterall, what Stephen did all those years ago, and he came away with good looking prints.

The comparison begins with the base 6000x4000pixel image unsharpened, followed by "Capture Sharpen" version.  I see a useful improvement in "sharpness."  The effect is rather dramatic, actually.  

The out of focus rendition wasn't much effected, though I did note higher contrast when using "Capture Sharpen" and perhaps a "grainier" feel to the out of focus region when pixel peeping. This would hold true for subsequent process variations, too.

Using the Gimp's "NoHalo" 9000x6000pixel up-sizing operation on an un-"Capture Sharpen"ed image I see overall smoothness in the image.  There is little to no objectionable noise and pixelation seems to be well under control.

Applying G'Mic "sharp" function called "Inverse Diffusion" to the base un-sharpened, up-sized image I see that things still look pretty nice.  Though I don't show it here, this result is "sharper" than using the Gimp's "Cubic" up-sizing operator and the Gimp's USM sharpener that I used in my earlier studies.

Considering the "NoHalo" up-sized "Capture Sharpen"ed image, I do have to say, that looks eminently printable straight away.  Artifacts introduced by up-sizing a "Capture Sharpen"ed image seem fairly well controlled, even though they are more evident than in the un-Capture Sharpened images.  We can see these artifact when "pixel-peeping", but they will be slightly masked in a final print.

The "Inverse Diffusion" sharpen operation applied to a "Capture Sharpen"ed image is simply too much.  Well, to my eyes, at least.  So to tame the overall effect down a bit, I put the "Inverse Diffusion" sharpened image in a layer and set the opacity to 40 percent.  This seemed to be a pretty good balance between too much noise with too many artifacts, and further increases in "sharpness."  Balanced in this way the image really "pops."

To this point in my investigations I feel the un-sharpened/NoHalo up-sized/Inverse Diffusion sharpened image is very nice and is probably quite printable as is.  

However, for the ultimate "pop" without the feeling of being "oversharpened", the Capture Sharpen/NoHalo up-size/Inverse Diffusion at 40percent opacity really rocks my boat.

I didn't stop there.

How it occurred to me I will never know, but I thought about having RawTherapee add just a hint of noise reduction early in the process to see if it had any effect on the intensity of the artifacts introduced by "Capture Sharpen" in an up-sized image.  

I'd recently come to understand the RawTherapee "Noise Reduction" operation can be very subtle when I want it to be.  It can also act like a heavy hammer when the ISO's are through the roof and the noise is so great I can't sleep at night.  No, I'd try to take a very soft hand to the low-ISO image noise to see if I could "knock off the edges" just a tiny, nearly un-noticable bit. 

I implemented the following process -

  • Import image to RawTherapee
  • "Capture Sharpen" image
  • "Noise Reduction" applied with minimal action on the sliders
  • Pass the "tif" image into the Gimp
  • Up-Size using "NoHalo"
  • G'Mic "Inverse Diffusion" sharpen image in a layer
  • Set sharpened image layer to 40percent opacity

Take a close look at the last two image sets and compare them with images further up the chart.  What do you see?  Not half bad, eh?

To encapsulate my current feelings of which would be "best" -

  • Quite "adequate" - RawTherapee un-sharpen base image/Gimp NoHalo upsize/G'Mic Inverse Diffusion sharpen

  • Amazing "pop" - Rawtherapee Capture Sharpen base image/Gimp NoHalo upsize/G'Mic Inverse Diffusion sharpen layer with opacity set to 40 percent

  • Pleasingly "luscious" - Rawtherapee Capture Sharpen + very subtle Noise Reduction of a base image/Gimp NoHalo up-sizing/G'Mic Inverse Diffusion sharpen layer with opacity set to 40 percent

After all is said and done... should I be able to take a 6000x4000 Sony sensor'd image, apply any of these three process versions, and come away with a beautifully printable image that is 48 inches long? ... maybe ... maybe yes...  quite possibly ...

No comments: