[jpzk]

The other thread began with that link...


Steve

Alright then...so I created a monster !!

[joergens.mi]

Yes, it's Rishi, how did you know?

Than we know the quality of this article.
He claims to be a scientist, but he is a shame for that profession - even if he has a graduation.

[MossyRocks]

I frequently use the technique discussed in the article but not to get higher resolution. I use it as a ND filter simulator but since I was doing the stacking anyway I take the extra resolution and extra bit depth in the output file as it does allow more freedom, even if I have never needed the extra resolution. The extra bit depth I have enjoyed having, especially with some of the night photos I have taken. For this I like the results but there I decided before I got the camera out what I was planning on doing and having experience working with this I know what kind movement artifacts I expect to see and if I don't want them in the picture I won't take them.

When I first read about the super resolution technique over a year ago I became fascinated with it and played around with trying to extract as much info out of bad cameras as possible (cellphones and the wife's cheap point and shoot), I even played around a little with it on film. My takeaway was that the biggest benefit is its ability to stamp out noise on cheap sensors, or get rid of grain in film pictures. The added resolution always seems a bit soft even if it was better than available up scaling methods. On a better camera the the gains aren't as great as and if you are chasing extra resolution you had better take a lot of pictures. This means get your self a really powerful computer with a ton of ram (which I have). Processing does take a while to align the images even on my computer, with a 2x upscale 40 image stack from a K-3 taking about 10 minutes, so I don't know where the hours of processing is coming from.

The inverse square law is the killer with this technique so 4 images cuts the noise down by half under ideal cases which is a huge gain but to get another 50% decrease in noise takes 16 images but there you are going for a very low noise image to an even lower noise image. At some point shortly there after you will push the noise below the 16bpc in the image so unless you want to go up 32bpc images it just isn't worth it on the noise end. You can still chase the extra resolution here but you really need a lot of images again as the inverse square law starts creeping in. So a good compromise may be stick around at a stack of 16 images (maybe a few more since we are working with less than ideal sensor positioning) and go for a 50% decrease in noise and a doubling of the resolution.

[monochrome]

Now that we know it’s another DPR Fail (to paraphrase the conclusion from Comment #7) can we just Ignore them and move along?

[joergens.mi]

I used this technique two, but to decrease noise. I did a free hand shot (16 pictures in a series) at ISO 2500 (resulting pseudo ISO 160) overlaying them by median and it worked fine. I did it because I had no tripod around and no time to go there a second time. ( Here is a result File:Lorettokapelle (Freiburg) jm62098.jpg - Wikimedia Commons ).

But when I look a the work I had to do with 16 pictures ACR (all together), loading into PS, putting into one picture as layers, setting to smart, making median, reducing to background exporting. roundabout 20 Minutes on 6 Core 4 GHZ machine with 32 GB RAM 36 MPix ist big even for such a system.

[Adam]

PS, it says produces unwanted processing artifacts. Too funny, not in any image I've ever taken, but hey. DPR says so so it must be true. The correct statement would be on some occasions PS produces unwanted artifacts.

This isn't really an issue any more because those artifacts can be detected and suppressed quite trivially (i.e. through the motion correction feature, which is always on in DPSR mode).

[Oakland Rob]

I'd say my experience with DPS images from the K-1ii is about the same as what DPR is reporting.

I've done in-camera JPEGs, DNGs and some brackets that I stacked in Photoshop. As well as changing the DPS DNG into a JPEG in-camera, and pulling out the four frames from the DPS DNG and then stacking those in Photoshop. And I'd say maybe the Photoshop ones sometimes looked better, but some of that might be down to processing and the fact that I can drop bad frames. But in truth I'm still not sure about DPS; seems sometimes I get considerably better images, sometimes meh. I'm still not sure why. And I dunno that it's worth having the camera locked up. Part of that is that the K-1ii gets great normal shots, and frankly I think the stacking helps more with smaller sensors and darker images with more noise, where (as in astro) the noise removal is more of a factor. It's really tough to tell the difference between a DPS and a regular shot on the camera's screen. Maybe if I used my loupe, I dunno. But hey, I rarely combine brackets for HDR purposes in-camera either. Or focus brackets (which I can do in camera with my Oly). MIght be more useful if you wanna a hi-res JPEG on the spot though.

I still think normal PS is better, and I do sometimes find that better than stacks in Photoshop. But that depends on the motion, whether RT or DCU can remove that effectively, etc. When PS is on, it seems to always come out better than a Photoshop stack (for me), but in iffy conditons, I like the control of using Photoshop.

[Rondec]

To me, this handheld technique is probably not so useful for "super resolution" although clearly it does work for that, but more to improve noise and color depth in situations where you have to shoot higher iso than you want -- say, a sunset and no tripod and you end up at iso 800. If you have a tripod, traditional pixel shift is definitely the way to go. If you don't have a tripod and still want to maximize your dynamic range and control noise, then DPS is definitely an option. Clearly, if there is a lot of wind, a bunch of subject movement you are just better off sticking with a single image.

I shoot a lot of pixel shift images (traditional) from a tripod and there are plenty of times that there's a decent breeze blowing and I get the images loaded onto Raw Therapee and check the motion mask and the whole image is basically green. In that situation, there is no benefit to using pixel shift and you just pick the best single frame and develop that. I'm sure the same is true for DPS. The camera is adjusting for camera movement and it can mask out a certain amount of subject movement, but if everything is moving, then there is simply no point to using it.