Interesting hobby, this photography.
Long ago, I came to realize that if a number of otherwise knowledgeable and objective people had serious disagreements on an issue, then either there was a misunderstanding about the premise, the details, semantics – or else it really just didn’t matter.
Previously I commented about differing workflows on an ISO 800 image of an outdoor scene. This time I selected another image and again varied the workflow: NeatImage, uprez, capture sharpen and output sharpen. In many of the tests, I just swapped the sequence of NeatImage and uprez. For good measure, I threw in the Fred Miranda 20D Resize. I tried NeatImage with a 20D profile from the NeatImage website as well as the autoprofiling feature; both the “remove all noise” and “default” filters were tried. For uprezzing, I used Photoshop bicubic smoother or the ACR upsampling. I tried a 4X uprez, a 2x uprez and a 1.75 uprez (maximum in ACR for the Canon 20D).
The new test image was an indoor portrait taken with a Canon 20D, RAW, ISO 400 and bounced, diffused fill flash mixed with sunlight. For most of the tests, I began with a severe crop to an eye and a bit of hair; this crop was a 214x148x16 bit image. Variations were output sharpened for monitor viewing and examined.
After making my conclusions based on the monitor view, I made three test prints and printed them on a Canon i9100 using 13x19 inch matte paper. The three prints were each a 1.75 uprez from the Canon 20D native size and printed at 300 dpi; that is, the full size print would have been 20.5 x 13.7 inches. Prints were made as follows: (1) ACR using 1.75X, NI auto profile, capture sharpen, print sharpen (2) ACR using 1.0X, NI using 20D profile, capture sharpen, Photoshop 1.75X using smoother, print sharpen and (3) ACR using 1.0X, Fred Miranda 20D Upsize routine to 1.75X. Instead of making three prints, I cut a center strip from each variation and made a composite without further upsizing. The composite was printed and examined.
It seemed to me that Method (1) above would produce the best print whereas I’d like to routinely use Method (2) and I was curious about Method (3).
The results? It doesn’t matter.
Well, actually it does matter, especially for onscreen viewing following large upsizing. In that case, with my “set of tools”, the best workflow is obviously uprez, apply NeatImage, capture sharpen, then output sharpen. For this workflow, it is necessary to make a NeatImage profile for the particular image being processed because the camera specific profile is no longer appropriate. Granted, images for viewing onscreen are not likely to have been significantly upsampled except in the case of severe cropping but that does happen.
On the other hand, the prints were about the same and large prints are probably the main reason for uprezzing. Methods 1 and 2 were nearly identical; most people could not tell the difference between them. The Fred Miranda variation showed just a bit more “texture” although very acceptable. I’d be happy with any of these prints.
Having gone through this exercise, when I make a large print from my 20D, I’ll be using Method (1): ACR using 1.75X, NI auto profile, capture sharpen, print sharpen. Why would others use variations of Method (2)? Well, Method (2) is a very rational workflow and produces good results. Actually, most of the time I’ll be using Method (2) and varying the output dpi to get the desired print size; Method (1) will be necessary only if the image has been extensively cropped.
Considering the many variations available in noise reduction, sharpening and upsizing routines, it is reasonable to me that noise reduction could be the first step in a workflow but this conclusion is not as obvious as might be first thought.
(I hope someone else will run a few tests, especially on some higher ISO images.)