SNS-HDR v1.x

Want to talk about SNS-HDR, or do you have suggestions, requests, or how-to questions? Post here!
Chcesz porozmawiać o SNS-HDR, masz sugestię, próśbę, lub jakieś pytania? Pisz tutaj!

Re: SNS-HDR v1.x

Postby Redcrown » Thu Apr 30, 2015 5:43 pm

[attachment=0]GhostStackPost.jpg[/attachment]The recent release of Lightroon 6 and ACR 9.0 included a "new" HDR function. Emphasis on "new" because while it looks like a port of the HDR Pro that has been in Photoshop for a while, it is not. There are significant differences. One of those is improved deghosting.

The following attached image set is a small 100% crop of a 3 bracket set. The exposures were at 1/200, 1/500, and 1/1250. All 3 originals are sharp, no movement blur. But due to a moderate wind, there is significant movement of the leaves and branches between exposures. A typical landscape ghosting problem.

The "Originals" frame was made by layering the 3 exposures and setting opacity of the top layers to 50%. It shows the actual movement of elements between shots. The other frames are labled by the HDR program used to create them. Default settings were used on all, no atttempt was made to match tones. This was a deghosting test only.

ACR-HDR (same as LR6) has 3 deghosting settings: low, medium, high. This test used medium. Oloneo has 2 settings. Method 2 was used because method method 1 failed miserably. Photomatix has a deghosting slider that goes from 0 to 100. At 50, Photomatix also failed miserably. At 100 is was much better and showed no artifacts. SNS has no deghosting variable. It's just on or off.

Observations: ACR-HDR did a very good job of deghosting. Best I've ever seen, in this test and in several others I did. Photomatix was a close second. Oloneo left a number of random ghosts. Not bad, but not useable without a lot of repair work. SNS was a distant last. Almost no ghost removal, and not useable at all.

SNS has been my favorite HDR program because I prefer its tonemapping, but it can't be used when even moderate ghosting occurs. The new ACR9/LR6 HDR function has now become a top competitor. Its tonemapping is more "natural" and is very close to SNS. It can't do the over-the-top, cartoon like images that Photomatix, Oloneo, and other HDR programs are famous for. But like most people, I don't care about that.

http://kellyphoto.smugmug.com/photos/i- ... bkNV63.jpg
Attachments
GhostStackPost.jpg
GhostStackPost.jpg (264.44 KiB) Viewed 17249 times
Redcrown
 
Posts: 30
Joined: Thu Apr 30, 2015 5:34 pm

Re: SNS-HDR v1.x

Postby Sebastian Nibisz » Thu Apr 30, 2015 6:04 pm

LR 6 remove ghosts well, but this method greatly increases the noise in darker areas adjacent to the light areas. I can implement a similar method in the SNS, it is not difficult.
User avatar
Sebastian Nibisz
Site Admin
 
Posts: 947
Joined: Tue Sep 13, 2011 8:51 am

Re: SNS-HDR v1.x

Postby JPS » Thu Apr 30, 2015 6:20 pm

Redcrown wrote:ACR-HDR (same as LR6) has 3 deghosting settings: low, medium, high. This test used medium. Oloneo has 2 settings. Method 2 was used because method method 1 failed miserably. Photomatix has a deghosting slider that goes from 0 to 100. At 50, Photomatix also failed miserably. At 100 is was much better and showed no artifacts. SNS has no deghosting variable. It's just on or off.

Sorry to hijack the thread, but I have a question about "ACR-HDR" as you call it.... Is this new ACR function implemented only in Lightroom 6, or in an ACR update in PS CS 6 ?

Cheers,
J-P.
Photo Galleries: http://www.pbase.com/scherrer - Spherical Panoramas (360x180°): http://www.viewat.org/?sec=pn&id_aut=2489
STUFF: Nikon D810, V1 + a few lens
User avatar
JPS
 
Posts: 26
Joined: Thu Mar 12, 2015 2:34 pm
Location: Brazil, on the beach

Re: SNS-HDR v1.x

Postby JPS » Thu Apr 30, 2015 7:50 pm

I found the answer to my question above !

It seems that the new HDR thing implemented in ACR will only exist in Lightroom 6 and Photoshop 6 CC !!! No way to have it as an update for the "normal" Photoshop 6 !!

Bad bad bad Adobe !!

I'm getting more and more pi$$ed-off with Adobe !

:-(
Photo Galleries: http://www.pbase.com/scherrer - Spherical Panoramas (360x180°): http://www.viewat.org/?sec=pn&id_aut=2489
STUFF: Nikon D810, V1 + a few lens
User avatar
JPS
 
Posts: 26
Joined: Thu Mar 12, 2015 2:34 pm
Location: Brazil, on the beach

Re: SNS-HDR v1.x

Postby Redcrown » Thu Apr 30, 2015 8:13 pm

Sebastian Nibisz wrote:LR 6 remove ghosts well, but this method greatly increases the noise in darker areas adjacent to the light areas. I can implement a similar method in the SNS, it is not difficult.


I understand. Adobe apparently chooses the darker of the ghosts and then raises the exposure of those pixels as needed, resulting in noise.

But in a merged image, I would much rather work on the noise than work on the ghosts.
Redcrown
 
Posts: 30
Joined: Thu Apr 30, 2015 5:34 pm

Re: SNS-HDR v1.x

Postby Erik Krause » Thu Apr 30, 2015 8:54 pm

Sebastian, if it really is not difficult to have ACR-like deghosting please implement urgently. Noise is no issue, at least not with a decent camera. Halos, like ACR creates and bad tonemapping like photomatix are much worse. I consider deghosting as the top primary feature that needs improvement in SNS, more than any other (since tonemapping is already superb).
Erik Krause
 
Posts: 38
Joined: Sun Jul 22, 2012 8:10 pm

Re: SNS-HDR v1.x

Postby Joghi » Thu Apr 30, 2015 9:28 pm

I´d also prefer better ghosting than less noise. Ghosting was often an issue in my landscape photos, so if you can handle it easily, it would be great.

greez Joghi
Joghi
 
Posts: 28
Joined: Tue Dec 09, 2014 12:50 pm

Re: SNS-HDR v1.x

Postby Bart_van_der_Wolf » Fri May 01, 2015 8:08 am

Erik Krause wrote:Sebastian, if it really is not difficult to have ACR-like deghosting please implement urgently. Noise is no issue, at least not with a decent camera. Halos, like ACR creates and bad tonemapping like photomatix are much worse. I consider deghosting as the top primary feature that needs improvement in SNS, more than any other (since tonemapping is already superb).


Hi Erik,

I slightly disagree. I don't think Sebastian should lower the quality to ACR standards, and that's also not how I think Sebastian would like to operate. Afterall, that's why SNS-HDR is better at tonemapping than other programs.

The method that ACR/LR uses in the more recent versions is based on work of Greg Ward Larson, a pioneer in HDR and author of many (ground breaking) articles (http://www.anyhere.com/gward/papers.html). If I'm not mistaken, it's based on using the median average of multiple images (after alignment and exposure equalization), which will exclude the pixels that changed more for the same pixel position than on the other images. That can be used as a mask to exclude those images. So if a feature didn't move (e.g. smooth sky, or a building), then it will be present at the same location in multiple (or a majority of) images, and one can pick the best exposed image from the sequence (or average a few) for lower noise.

For features that did move, they will more likely have a non-median value in the least common image(s). However, it then becomes harder to pick the right one. Like in the moving branches image example, it is not obvious for a computer if those are the same branches, or different ones, at a different location. This is where manual intervention (painting out one or more ghosts), or plain guessing comes in, unless one creates a much more sophisticated pattern recognition into the software. If a feature is similar but displaced/rotated a bit and maybe warped a bit, then one can pick the better exposed version and mask out the other replicas.

But I'm not so sure that that is easy to do with how SNS works internally. It looks like SNS builds a single image from multiple exposures, and then does its magic on that composite, but the anti-ghosting needs to take place before we can even see the image composite. So that would mean that every time that some de-ghosting is done, the entire image sequence needs to be re-imported/re-calculated and composited to a single version for tonemapping.

I hope I'm wrong and it's easier for Sebastian to implement than I estimate. After all, he knows the inner workings of SNS-HDR best. I do agree that for images with movement, good de-ghosting is very important, so it could be another area where SNS-HDR excells and beats the alternatives.

Cheers,
Bart

P.S. Here are some documents that may help in finding an efficient automatic method:
http://scholar.google.com/scholar?q=A+robust+and+fast+anti-ghosting+algorithm+for+high+dynamic+range+imaging&hl=en&btnG=Search
https://people.csail.mit.edu/kapu/papers/JHu_OGallo_KPulli_XSun_CVPR2013.pdf
https://www.cl.cam.ac.uk/~rkm38/pdfs/hadziabdic13cda.pdf
http://web.cs.hacettepe.edu.tr/~aykut/papers/hdr-deghosting-star.pdf
Last edited by Bart_van_der_Wolf on Tue Nov 08, 2016 10:13 am, edited 1 time in total.
Bart_van_der_Wolf
 
Posts: 10
Joined: Thu Oct 06, 2011 10:42 am
Location: The Netherlands

Re: SNS-HDR v1.x

Postby Erik Krause » Sat May 02, 2015 5:47 pm

Of course I don't want SNS to lower it's quality to ACR. But a state-of-the-art deghosting (but maintaining the tonemapping quality of course) would make SNS to the single very best HDR program.
Erik Krause
 
Posts: 38
Joined: Sun Jul 22, 2012 8:10 pm

Re: SNS-HDR v1.x

Postby Redcrown » Tue May 05, 2015 6:03 am

I need to take back most of what I said in my earlier post about deghosting of SNS, ACR, and Photomatix. And acknowledge that Sebastian is absolutely right about noise "greatly" increasing in areas where ACR removes ghosts.

I set up a controlled studio environment with a high dynamic range scene (10 stops) where I could move objects slightly between exposures to simulate ghosting. I used small, medium and large objects, moving them in both shadow, midtone, and highlight areas.

The results were horrible. Sebastian was underestimating the noise problem when he said it "greatly" increased. It takes a word greater than "greatly" to describe it. For an example, take a medium ISO (800) raw image with a deep shadow (RGB values around 10 to 15) into ACR. With no noise reduction, move the Shadows to +100, move the exposure to +2. What you see is ugly and what you get when ACR removes ghosts is even uglier than that.

The results are not bad when the ghosting object and its background are both centered around midtones. But if either the object or its background (especially the background) is in the deepest shadows, then the results are extremely bad. Far too bad to even think about using noise reduction to fix.

The bigger the object and the greater the movement, the worse the results. I think that's why I didn't notice any problems in my example of very small and dark tree branches moving against a bright sky. But if a big man is walking in the shadow of a building, there is no hope.
Redcrown
 
Posts: 30
Joined: Thu Apr 30, 2015 5:34 pm

PreviousNext

Return to Discussion & Questions / Dyskusja i pytania

Who is online

Users browsing this forum: Bing [Bot] and 2 guests

cron