[Feature request] SuperResolution


(Helder Vicente) #1

Hello everyone.
I have been using HDRMerge for a while, and I’m really happy with it(great results).

There is one feature that I would really like to see implemented on it: SuperResolution

Context:
So far I have been using the following workflow:

  • Take a burst shot (4 photos with different exposures ev -1/ 0 /+1 /+2)
  • Use HDRMerge to blend the pairs of raw photos so that i end up with 2 different raw files (ev -1 with ev+1 and ev0 with +2)
  • Resulting 2 different raw files from hdr merge
  • Import them into photoshop and do the SuperResolution workflow (upscale 200%, align, median stack blend)

This workflow has less noise and comparable detail to a pure SuperResolution performed exclusively in photoshop(with the 4 raw files taken).

Problem:

  • have to jump to another program to create a SuperResolution shot.
  • End up with a PNG file after exporting from photoshop.

Solution
Integrate SuperResolution processing on the HDRMerge options window (when it shows the added files ) with options for:

  • Image size percent increase (for example: 150%, 200%, 250%…)
  • Resample methods available

The SuperResolution would be performed on the RAW files, and then the HDRMerge would take those files and process them as usual.

Intended result
To achieve a SuperResolution HDR shot from different exposures combined in a single RAW file output


HDRMerge status update
HDRMerge status update
(Eduardo Bohoyo) #2

Hello, Vicente:

Aside from your request, I admit I do not get the advantage of two pairs of HDR mixed afterward vs one HDR, due to my poor knowledge. Would you mind explain me the foundation of this technique in more detail, o could you link me to some sort of initiation pages? I would appreciate it so much.

Thanks in advance.


(Helder Vicente) #3

Hello Eduardo,
the mix of images(using a median stack) should happen after they are resampled to a higher resolution.

The SuperResolution workflow is really impressive because:

  • It allows you to use hand held shots (for pixel shifting)
  • You can achieve higher details with a simple burst shot

Digital imaging allows us to have better results than our hardware can achieve.

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


(Helder Vicente) #4

Using the previously pointed out workflow, here are the results (500% zoom on a 7378x4924 file):

The image in the middle is using the HDRMerge with SuperResolution, it boasts the same detail and slightly less noise(smoother transitions) than pure photoshop.

I believe that if HDRMerge will adopt SuperResolution the results can be even better.


#5

Interesting, I’ve read about the technique and always wanted to try it.


#6

I never got any results that were better than the input images. Maybe I should give it another try.


(Morgan Hardwood) #7

Personal thoughts on the subject: I have followed the “superresolution” posts here and the conclusion I drew is that it leads to poor quality results and so is a waste of time. HDRMerge is a neat and small program which does one thing and it does it well. I would classify adding this sort of feature as bloat.

Median stacking is a separate and valid technique though with extremely narrow applicability.


(Helder Vicente) #8

@Morgan_Hardwood HDRMerge does it very well indeed.

The median stacking with pixel shifting (hand held shot) - SuperResolution - gives great results in terms of detail. It’s perfect for landscapes and shots without moving subjects.
Plus, there isn’t any app at the moment that takes several RAW files and upscales them to a RAW output.

That’s why i believe that having the option in HDRMerge would be great.
I previously updated a result of a very simple edit with great results.

Could you point out any examples on why you believe that SuperResolution leads to poor quality results?


#9

I’ve tried superresolution and one big issue is trees. Trees have branch intersections that move dramatically when the branches themselves move very slightly, and that really screws up stacking. Plus they sway as a whole a surprising amount. You’ve gotta have a perfectly still day to make much of superresolution from anything with trees outdoors.


(Eduardo Bohoyo) #10

I’d wish give Rawtherapee a try with your four NEF and my custom DNG output. Could you upload them?


(Morgan Hardwood) #11

I could have worded that better, let me clarify:

  1. “SuperResolution” sounds like a thing, but it isn’t. There are many techniques for increasing resolution, very different to each other (1) (2) (3) etc. In a technical discussion, something which could mean anything means nothing.
  2. You’re asking for two things: upscaling and subsequent stacking using a median or linear blend.
  3. Those things require perfectly aligned images. Not easy with various raw formats.
  4. Usability of such a feature is severely limited: the scene must be static, and if raw image alignment is not implemented then the shots would also need to be perfectly registered.
  5. Feature requests for blending and alignment have been open for ages, no one is working on them.
  6. HDRMerge has two developers, both heavily involved in other things. I don’t count myself, as I just fixed one bug and will update its website.

(Helder Vicente) #12

Thank you for the detailed explanation.

I thought HDRMerge already did the alignment properly?
One more question, does HDRMerge perform any sort of smoothing/noise reduction to the output?


(Helder Vicente) #13

Hello Eduardo,
here is the link with the samples I used to show the results:


(David Oliver) #14

Separate from the discussion of whether or not “SuperResolution” can/should be included in HDRMerge, here’s my Pentax K-3 II pixel shift test.

I used Pentax Digital Camera Utility (PDCU) 5.4. No lens correction or other changes were applied. For the non-pixel shift shot, I used PDCU’s standard sharpening mode (the fine sharpening mode wasn’t helping in the same way it does with pixel shift shots and seemed to be having an undesirable effect), and applied as much of this standard sharpening as I could before angled edges got very obviously pixelly/jaggy.

Whole-shot preview, just for context:

1:1 crops (you’ll need to click to zoom to show the full difference as these images are wider than the content column here in Discourse):

My original thread at Pentax Forums is here.

SuperResolution/pixel shift as I think of it doesn’t result in final images which have greater pixel dimensions, but images which are of the same pixel dimensions but with greater definition and less noise due to the combining of multiple shots. I’m not sure why upscaling would be necessary/desired? Is interpolating (substantially more) useful in that workflow in addition to merely combining and averaging shots at their original PPI?


#15

Olympus uses more than just 4 shots and manages to result in an image with more pixels than the original.


(Helder Vicente) #16

Thank you for the excellent images to illustrate the results of pixel shifting and stacking.
From my experience/tests using upscaled images increases the definition even further.


(Morgan Hardwood) #17

@DavidOliver I love my camera, but damn that’s a nice feature to have! What are the minuses of enabling it, how fast does it do the pixel shifting? How does it impact exposure time?


(Helder Vicente) #18

Even without a pixelshift enabled camera, we can get similar results with hand held shots on a normal camera.
We just have to use a burst shot, align the stack and median blend to get the improved detail.
That’s why a SuperResolution workflow is something very interesting to bring to HDRMerge.


(David Oliver) #19

Minuses include:

  • four separate exposures, quadrupling overall exposure time (not good for moving subjects)
  • currently, it complicates processing as the RAW processor needs to be able to process the resultant single Pentax RAW file, which includes four separate images. darktable doesn’t do this, so I would have to make a TIFF using either Pentax’s RAW processor or a modded version of dcraw, both of which can help with artefacts caused by movement between the individual shots, for importing into darktable/GIMP/whatever.

Doing it more manually, as @Helder.Vicente describes, avoids the non-standard four-in-one RAW file problem, of course.


(Pat David) #20

Why go into Ps when there’s perfectly good methods with Free Software? :slight_smile:

This could easily be scripted to automate it, and can produce a tiff file as output if you’d like.

I think they’re using up to 8 shots right now in the E-M1mk2.