John Sheehy wrote in post #19417495
There are a lot of poor processing options out there. One of the biggest problems is the fact that non-weighted downsampling methods, "Nearest Neighbor" being the worse of them by far, were program defaults in the past when computers were very slow, and they should have been dumped as defaults when computers got much faster at downsampling, but there is an appeal to NN to many people because when you have an image that does not have a lot of noise, NN can make the results seem "sharper", even though it is bogus sharpness due to aliasing. When someone is using NN and then starts working with noisier images, then NN greatly increases image-level noise the more you downsample an image. The noise in a 32.5MP image downsampled to 0.5MP is similar to that from a sensor with 1/65 the area of the actual APS-C sensor, about the size of a 12.9x crop from a FF camera, with 1/166 its area, and this noise isn't smoothed at the pixel level as happens to some degree in demosaicing; it is maximally sharp noise, impossible to ignore.
So, how many people when buying a higher-MP camera don't even realize that using NN for downsampling adds more noise that really isn't there, the greater the downsampling ratio?
If one is intending to downsample an image to show it at a small size on a coarse-pixel display, early sharpening is also your enemy, if done before the downsampling. Never sharpen anything until it is at its final pixel dimensions, and if you are going to downsample, you should probably hit the full res image with heavy NR or dumb median filtering of details that are too small to survive the downsample, first. I am talking about NR and median filters because they don't just blur meaningless dot outliers; they actually REMOVE them. Just blurring them before downsampling would maintain them after the downsample.
John, I always appreciate your detailed postings about resolution and noise. Even when I don't fully understand the underlying subtleties, I find your postings very informative.
The JPEG export in ON1 only has two selections for resize method, bicubic and "Resize AI". Nearest neighbor is not an option. Beyond that, the only settings it has are for quality (default = 90%) and export sharpening, which I have disabled. The export sharpening selections are screen, matte paper, and glossy paper, with high and low sharpening for each.
Your point about downsampling is also salient. As I mentioned earlier, I've begun using ON1's built-in de-noise function (NoNoise AI), and I do this as the final step before export, after I have made all other adjustments. While not a downsampling, it seems to clean up the noise enough in the images that the built-in JPEG export generally does an acceptable job. Arguably this indicates my previous problems with JPEG export were due to my own process, but GIMP doesn't seem to have any issues with the non-NR'd images in the full-size TIFFs.
The JPEG export is only one of my reasons for switching back to LR. With the 32.5 MP images from the R7, I'd like to go to a CRAW-based workflow, but ON1 doesn't support CRAW. When I emailed to ask if they planned to add it, I was told that there were "no current plans" for it. The rep did direct me to a site where I could submit new feature ideas that would be "voted on by the community". Thanks but no thanks.
There's also a lot more third-party and free educational material available for LR, whereas the main source of ON1 education is ON1 themselves (and you need a separate subscription for most of it).
