Living where I do, I _know_ very much how fortunate I am to live in peace. There is mental space and physical safety to do the things I want. Like write these little blog entries. Not everyone has this option these days. I wish peace for everyone.
-------------------------The distance from my first post on "SuperResolution" to the present is rather long and the path is fairly twisted.
In this latest series of posts I explore using the Gimp's new(?)/recent(?) NoHalo Image Scale up-sizing function.
Recall from Graphic Design that -
In my last post I looked at a black and white synthetic image and the various effects of up-sizing and UnSharp Mask sharpening. I found that NoHalo is sharper than Cubic interpolation and had none of the artifacts that the Cubic function did. And after applying a 1 pixel USM to the NoHalo interpolated image, things, frankly, looked quite good.
Test Image -
For this blog entry we will consider the NoHalo interpolation using my current favorite "test" target, a copy of the French Canard Enchaine newspaper. It's a great read, if you're ever interested in French politics and all the in's and out's of what it's like to try to govern a First World nation.
To make things more "interesting" I will present two comparisons, side by side (as seen below).
The left hand image was taken using a pretty Nikon Nikkor-P 105mm f/2.5 shot wide open at f/2.5. I felt this might be a good "test" to see how the slight softness seen wide open behaved when "Capture Sharpened" and then up-sized.
The right hand image was created using the very same lens, but this time shot at f/8. I felt it could be interesting to see what might happened when I up-sized a demonstrably pixel-sharp base image, and even to compare this against f/2.5 images.
[As always, click on one or both of the images and open at 100percent to pixel-peep]
Comments -
Note: Again, keep in mind that we are _not_ adding detail to an image by
performing an up-size of this kind. You can see this by my use of the word "up-size" in place of "Up-Rez." If the information isn't there in the base file,
then it certainly will not be there in the interpolated larger image, either. So saying any of this is found under the heading of "SuperResolution" can be mis-leading. In general, I've found that if I _really_ need extremely high resolution that it's better to shoot overlapping scene sections and then to stitch them together into a larger image.
We can easily see that RawTherapee's "Capture Sharpen" function works the treat on a base image. In both the f/2.5 and f/8 cases, the 100 percent pixel-peep'd base images look and feel sharper after "Capture Sharpen." Additionally, there's not much difference between the "Capture Sharpen" base images shot at the two apertures, either.
Starting with the 1.5x linear Cubic Up-Rez'd images, the edges are slightly "soft" after the interpolation to 9000 pixels. Then, with a 1 pixel wide UnSharp Mask applied, it appears to me as if the image "cleans up" fairly nicely and is quite usable.
With the 2x linear Cubic up-sized images we begin to see the phenomenon that started me down the path of considering other image scaling interpolation algorithms. The edges of the transition zones look slightly "soft." This remains true after applying a 1 pixel USM as well.
Looking at NoHalo straight out of the interpolation step, the 2x linear up-sized image looks sharper than the Cubic Up-Rez. This, by going from a 6,000 x 4000 pixel image (24mpixel sensor) to a 12,000 x 8,000 pixel image size. Crazy, isn't it?
Adding a 1 pixel USM to the NoHalo up-sized image seems to help it go from looking good to looking great. And there are no obvious artifacts.
This is what I was looking for when I started down the path of increasing an image's file size.
No comments:
Post a Comment