Is it possible to combine two versions of the same image to improve its resolution?

[RauPP]

I'm often scanning images at very high resolutions, and even if you set up and clean everything carefully, very tiny disruptive factors like a mote of dust, a ream on the scanner surface, an ever-so-slight hiccup of the scanner sensor that materialises in a blur or 2,3 missing lines of pixels and the like can slip through and remain visible in the final image. Usually it's not a big deal (nothing the stamp tools can't fix), and if it is, I do another scan and clone a portion of the new scan into the affected area of the old one.

But it got me thinking back to the techniques I used in the past to develop digital photographs, especially HDR photography. With certain kinds of HDR photography, you combine multiple images of the same motif, taken with different exposure times, into one image with higher dynamic range than any of the single images. Photoshop recalculates every single pixel of the image for that.

Now, it might be a far out idea, but I was wondering if Photoshop supports any functions to automatically combine two scans of the same image into one of better quality (as opposed to HDR on the one hand, and stitching together scans/photos of different image sections together to a larger panorama with Photomerge on the other hand). That could help remove particles that are not part of the original while preserving small details that are (like film grain in old photographs) and maybe even improve resolution beyond what a scanner can deliver at max res - at least that was my train of thought. Do you know what I mean? Does something like that exist?

[JJMack]

I do not think you mean resolution you mean improve image quality.  The answer is yes you can improve quality by blending versions. Of the same image.  To be able to Blend or blend and composite versions all image must have the same number of pixels image the image need to be the same size.  You can not do things like the with diffent size images.  Stack Mode blending can improve the image pixels quality.

[D Fosse]

That's an interesting question, and one I was asking myself some time ago. I don't know the answer, but intuitively I have a feeling it should be at least theoretically possible. But don't expect it to know dust from detail...

A problem with flatbed scanners (aside from real resolution way below the advertised one) is what you could call random "ripple" distortion - things aren't always in the exact same spot. Align for one part, and another goes out of alignment. So that might defeat the whole exercise.

If you have time for experiments, I would start with File > Scripts > Statistics. There are some interesting blend functions there. I've never had time to explore them.

[Semaphoric]

Maybe you could do something with an Image Stack. I've often wondered if one could reduce heat shimmer using multiple photographs that way.

[Chuck Uebele]

There is a process called "Drizzle" that is used to increase resolution on images - mainly for astrophotography, but also for forensics. It averages several different images. The key is to have them slightly misaligned, so the ripple effect with the scanner, might actually help in this regard, but I would move the copy in the scanner several times. This link explains it better:

http://www.stark-labs.com/craig/articles/assets/Drizzle_API.pdf

[davescm]

There is a technique in the link below which discusses improving resolution from DSLR in Photoshop through multiple images. Similarly to the Drizzle mentioned by Chuck, it works due to slight misalignment between each image, but the method in the link relies only on Photoshop functions to build the final image.

http://petapixel.com/2015/02/21/a-practical-guide-to-creating-superresolution-photos-with-photoshop/

It should be worth a try - particularly as it costs nothing.

Dave

[RauPP]

Fascinating stuff, guys. I was intrigued and had to try it out. After a couple of test runs that didn't bring any additional detail (and often even made things worse) I got the hang of it. In general I like the results, but they are not invariably better than a single high definition scan.

Look, for example, at this detail from my current main project, a restauration. The original is only about 12 mm high and 9 mm wide:

Nähen-Obstkram Detail - 2400 SPI Scan:

https://picload.org/image/raldpcwd/naehen-obstkramdetail-2400spis.jpg

Nähen-Obstkram Detail - 10 x 2400 SPI Scans combined to hyperresolution image:

https://picload.org/image/raldpcwg/naehen-obstkramdetail-10x2400s.jpg

There is an increase in detail in certain areas for sure. Especially the diagonal black strokes in the lady's hat are suffering from less aliasing in the hyperresolution image.

In other areas the effect doesn't seem to go beyond a glorified Blur filter: The cheeks, the paper structure itself and so on. Maybe you could even argue that in areas of more spacious color application the hyperresolved image (at least in this sample) is encountering a certain loss of detail.

Which brings me to another question: Are there any ways to quantify or measure the level of detail in an image? Of course I don't just mean pixel resolution. Sometimes two versions of an image (say, two different scans, or images taken by different cameras) can be so close in quality that your eyes can not really make out which one is the more detailed anymore, at least not without comparing many, many reference points. Have you guys ever seen/heard of/used a software or tool that can basically compare two images and say with a high degree of reliability "yeah, this one's more detailed"?