Let's Talk About...sharpening

The Number One reason I threw away Photoshop in favor of the Gimp is because its' Sharpening algorithms blow Photoshop away. 40% in the Gimp is 15% in Photoshop, and I've never had to use Unsharp Mask in the Gimp. With that said, I walked away from Photoshop ten years ago and things might have changed.

I know this is redundant (more for the audience), but you Sharpen at full image size, right?

Well, that kind of goes without saying....

Downsized images have fewer pixels. Why would you want to act on fewer pixels with a process that moves/changes pixels? Give the program all the granularity you can. Resizing is always the very last step for me.

I save the original, a full-size completely processed image, and a downsized image for every shot I keep. You never know what size you're going to want to post the processed image down the road.

To give you an idea of my level of paranoia, I never save whatever the final image is on my screen. I exit the program with that image in place, and refuse to save it when it prompts. I don't remember the last time I used "Save" as opposed to "Save As" (now "Export As" in the Gimp because it defaults to .xcf) - 1980's maybe, when I was in the sign industry? Even my in-progress backups are Save As.

Yeah, you should definitely do your image processing on the full resolution image - you could crop first that is a different matter. Sharpening is a very fast process. You will see many fewer "artifacts" in your end if you process first then resize.

My first coin (and other) photos with a DSLR were taken in 2004, with a 6MP Canon 10D camera. Currently, I use an 18MP? Canon 550D/T2i,which is not the latest and greatest, of course. Once the photos are reduced and JPEG converted for uploading to Flickr or CCF, you can't see much difference between the images from the 10D or 550D.

Of course, for closeup and macro shooting (coins in particular), the newer camera has live view, and improved tethering support. It does video for YouTube, too. My next camera will presumably have 24MP or more, since that's what the new cameras seem to have. As long as the per-pixel read noise scales down with the pixel size and well capacity, and the number of stored electrons per square millimeter doesn't suffer, the more pixels the merrier. The main downside is larger raw files and TIFFs that take up more disk space. Fortunately, disks are getting bigger too.

That's my assumption, but I don't have your ability to look at things naturally from an engineering standpoint. There's no denying the quality of your work, so rather than debate it with words here I'm just gonna have to see how I like your way.

Most of the shots I publish are either 0.64MP (800x800) or 1.1MP (1296x864, which is T2i/4 or XS/3). The original full-size images are quite large and unpublishable in any practical sense, unless a large print is being made, which I rarely do. The downsizing process gives "artificial sharpening" since it uses detail from a group of pixels to create a single pixel during downsizing, yet that process is not perfect and requires some further help to look its best.

This artificial sharpening concept is actually very similar to what happens with diffraction. Although they are completely different processes, it is instructive to think of them from the same perspective of how they affect sharpness. Diffraction due to small apertures in effect blurs detail across adjacent pixels. You can think of a larger aperture as doing the opposite by refocusing the detail onto the individual pixels. An even larger aperture focuses the detail to sub-pixel regions. But below a certain threshold you don't get any more detail in the image because the pixels are larger than the detail available from the lens at the larger aperture.

This same thing happens with downsizing of an image. The pixels of a high-MP sensor are analogous to the sub-pixel detail you get from large apertures. By downsizing, you are effectively increasing the aperture size at the downsized resolution. 2x downsizing is like opening your aperture 2 stops. 4x is like opening 4 stops.

This sounds great, but if the image is reasonably sharp at the pixel level, how much does this artificial sharpening actually help? In general I can't tell the difference between downsizing 3x vs 4x, which is why I have been recommending the Rebel XS camera rather than a more modern, higher MP count camera like a TXi. Once you do at least a 2x downsize, there is seemingly no further benefit, assuming the original full-size image was not heavily diffraction-limited. If it was, then additional downsizing might further improve the sharpness.

A big reason for the improved sharpness on downsizing, beyond any diffraction reduction effect, is that the pixel-level sharpness on most sensors is not optimum. Pixels are not perfect, and the imperfections (due to demosaicing errors, AA filter blurring, etc) immediately limit 100% pixel-level sharpness to a certain level. Downsizing by at least 2x utilizes adjacent pixels to compose the final image and reduces/eliminates the sensor artifacts.

All this said, much of the work I've done in the last few years has been devoted to getting maximum sharpness and detail at the pixel level, which allows 100% crops to show well. You might remember when I first started posting on this forum that my stated goal was to be able to take a single shot of a Lincoln Cent such that I could dig down into and view variety details, surface condition, marks and such. I so far have not succeeded to my satisfaction with optics alone, as the sharpest lenses I own (the sharpest available!) are seemingly insufficient for this purpose.

YET I see images that show more detailed 100% pixel level images than I have achieved. The images from the D810 have this quality, and I expect those from the latest Canons to as well. Both companies tout their latest image engines as being able to extract sharper detail at pixel level. It was my intent for this post to see if there are any techniques out there that can be used on older cameras to do the same thing.

Ultimately though, even if a sharp pixel-level image is achievable through better sensors, better optics, better demosaicing, or better sharpening algorithms, I would still stick with my described workflow, though I will be doing some additional tests to see what level of pre-sharpening could help with the final downsized image.

How about doing sharpening during the downsizing (and no sharpening before or after)? This would be the Photoshop "Bicubic Sharper (best for reduction)" method on the image size menu.