HDR Frustration

[Robert Boire]

Hi,

I have just started experimenting with HDR and am somewhat disappointed with the results so far.

I have taken a number of high contrast, landscape scenes in various locations. The composition of these scenes are basically simlar a bright background (top) half (sky, snowy mountains, glaciers ) and a subdued (not dark) foreground (bottom) half (forests, lakes etc). I have bracketed the exposures by 1 stop. In general the middle ("correct") exposure has good contrast, but of course has an underexposed foreground. The + 1 stop has about the right foreground, overexposed background, the -1 stop is generally too dark, though the background is arguably more dramatic. (My meter seems to indicate the range is more likely +/-  2 stops, but this tends to make the defects of the individual exposures even worse).

I am generating the HDR's using the native photoshop tools.  The results show approximately correct exposures, though the foreground tends to be washed out. Above all most (not all) of the HDRs are flat with the contrast largely lost. I have tried adjusting exposure and gamma and the various conversion options when converting to 16 bit, but for the most parts the result are disappointing. Except in the most extreme cases, I get better results (but not great results) manipulating the middle exposure.

Any advice out there? Am I doing something wrong?

PS:

I do not understand how the HDR algorithm combines the shots or decides which part of each shot it keeps, but this seems to be part of the problem. Are there HDR applications that allow more flexibility over which part of a shot are included in the composite?I have had some better results with digital GND filtering techniques, but this is extremely labor intensive and requires exacting precision. Are there "automated" tools out there for digital GND?

Thanks

[Steve Weldon]

Hi,

I have just started experimenting with HDR and am somewhat disappointed with the results so far.

I have taken a number of high contrast, landscape scenes in various locations. The composition of these scenes are basically simlar a bright background (top) half (sky, snowy mountains, glaciers ) and a subdued (not dark) foreground (bottom) half (forests, lakes etc). I have bracketed the exposures by 1 stop. In general the middle ("correct") exposure has good contrast, but of course has an underexposed foreground. The + 1 stop has about the right foreground, overexposed background, the -1 stop is generally too dark, though the background is arguably more dramatic. (My meter seems to indicate the range is more likely +/-  2 stops, but this tends to make the defects of the individual exposures even worse).

I am generating the HDR's using the native photoshop tools.  The results show approximately correct exposures, though the foreground tends to be washed out. Above all most (not all) of the HDRs are flat with the contrast largely lost. I have tried adjusting exposure and gamma and the various conversion options when converting to 16 bit, but for the most parts the result are disappointing. Except in the most extreme cases, I get better results (but not great results) manipulating the middle exposure.

Any advice out there? Am I doing something wrong?

PS:

I do not understand how the HDR algorithm combines the shots or decides which part of each shot it keeps, but this seems to be part of the problem. Are there HDR applications that allow more flexibility over which part of a shot are included in the composite?I have had some better results with digital GND filtering techniques, but this is extremely labor intensive and requires exacting precision. Are there "automated" tools out there for digital GND?

Thanks

Sample images would help.. a lot.

I find many people when first starting try to apply HDR techniques to scenes not well suited to HDR, or if they do they get the exposure offsets way off.   So it helps to have a peek.

[wolfnowl]

It depends on what you're talking about with respect to the term 'HDR'.  There's a very good article on this site for example about HDR that has nothing to do with the (often) overused, overprocessed, grungy look that most people associated with HDR: "HDR Plea".  Vincent Versace has taken to calling the type of image that Alexandre writes about 'eXDR' images for 'extended dynamic range'.

Anyway, Terry White has some good tutorials on using Lightroom & Photoshop to do the 'HDR look' thing - available on the Adobe TV site.  Photomatix is probably the most used HDR software out there, but others are welcome to disagree!

Mike.

[milt]

You should also stop thinking in terms of what "part" of each photo is used in the result.  That's not the way it works.  If you shot 5 bracketed shots, the HDR software looks at the same pixel in each of those five shots, takes those five values and applies a formula specific to the HDR software to come with a value for that pixel.  That value is expressed using an extended representation (usually one 32-bit floating point number per channel).

--Milt--

[Steve Weldon]

It depends on what you're talking about with respect to the term 'HDR'.  There's a very good article on this site for example about HDR that has nothing to do with the (often) overused, overprocessed, grungy look that most people associated with HDR: "HDR Plea".  Vincent Versace has taken to calling the type of image that Alexandre writes about 'eXDR' images for 'extended dynamic range'.

Anyway, Terry White has some good tutorials on using Lightroom & Photoshop to do the 'HDR look' thing - available on the Adobe TV site.  Photomatix is probably the most used HDR software out there, but others are welcome to disagree!

Mike.

Thinking about what Milt said.. Photomatix also allows blending.. and for some shots blending might be preferable.

I've had a bit of fun this weekend with the Sony NEX-5 in the "auto-hdr" mode.. (I'll be shooting it in jpeg until LR3/ACR/C1pro supports it in raw) and its amazing how well it does.  At 7fps, especially being able to use the articulated LCD as a waist level finder.. it adds another dimension to a pocket portable.  I spend half my time with the NEX-5 deciding if I want a HDR, expanded DR, a HD movie, sweep pano, bracketed exposures, or just a regular old image.. Once I figure out what I want to do with it.. then its easy from there..

[BobFisher]

Welcome to the wonderful world of HDR.  It can be a frustrating experience in the early stages.

What you're experiencing is common.  Without seeing sample images it's difficult to give any concrete ideas of how to go about addressing it.  

What version of Photoshop are you using?  The tools up to CS4 weren't great.  CS5 is much better with the addition of HDR Pro.  There are still ways to get good results in prior versions and you may want to look at what's called 'soft tonemapping'.  

Toss some samples up for folks to have a look at.  Both input images and untonemapped merged images (drop to 16 bit but do no tonemapping).

[milt]

I probably should have mentioned that the process of my previous post is step 1 of a 2 step process.  Once you get the high-dynamic-range representation, you then have to map it down to the limited dynamic range of our conventional image file formats (and our eyes, and our monitors, and our printing mediums).  This down-mapping is where there is a lot of subjectivity and there are a lot of choices to be made.  In Photomatix there are two modes for doing all this.  In mode 1 you explicitly do the 1st step ("Generate HDR Image") and the 2nd step ("Tone Mapping").  In mode 2 ("Exposure Fusion") they hide the fact that it is a two step process, doing it all behind the scenes.  Mode 2 is simpler, less powerful, and occasionally better. (I think this is what Steve calls "blending".) Mode 1 is complex, powerful, and, if done properly, usually better.  In addition, mode 1 is probably the easiest to misuse image manipulation software on the planet (it is the origin of that garish "HDR-effect" you see so often).

BTW, "dropping to 16 bit" is of course a form of tone mapping -- its just one where you don't get any say in what choices are made.

--Milt--

[BobFisher]

I probably should have mentioned that the process of my previous post is step 1 of a 2 step process.  Once you get the high-dynamic-range representation, you then have to map it down to the limited dynamic range of our conventional image file formats (and our eyes, and our monitors, and our printing mediums).  This down-mapping is where there is a lot of subjectivity and there are a lot of choices to be made.  In Photomatix there are two modes for doing all this.  In mode 1 you explicitly do the 1st step ("Generate HDR Image") and the 2nd step ("Tone Mapping").  In mode 2 ("Exposure Fusion") they hide the fact that it is a two step process, doing it all behind the scenes.  Mode 2 is simpler, less powerful, and occasionally better. (I think this is what Steve calls "blending".) Mode 1 is complex, powerful, and, if done properly, usually better.  In addition, mode 1 is probably the easiest to misuse image manipulation software on the planet (it is the origin of that garish "HDR-effect" you see so often).

PM's Exposure Fusion isn't a 32 bit process.  It's an exposure blending algorithm similar to TuFuse or EnFuse.

BTW, "dropping to 16 bit" is of course a form of tone mapping -- its just one where you don't get any say in what choices are made.

--Milt--

Fair enough.  Drop to 16 bit but do no 'additional' tonemapping.

[Robert Boire]

You should also stop thinking in terms of what "part" of each photo is used in the result.  That's not the way it works.  If you shot 5 bracketed shots, the HDR software looks at the same pixel in each of those five shots, takes those five values and applies a formula specific to the HDR software to come with a value for that pixel.  That value is expressed using an extended representation (usually one 32-bit floating point number per channel).

--Milt--

I guess I understand that to some extent, which is maybe part of my frustration. The question is has does the software come up with the value (some sort of averaging?) for the pixel. It seems to me that there are parts in any of the bracketed shots that clearly should not be considered since they will be vastly over or under exposed.

For those that asked I am using CS4.

This down-mapping is where there is a lot of subjectivity and there are a lot of choices to be made.

I understand that a lot of the art is in the down conversion, I just do not seem to be able to get to where I want to go. However I am admittedly not an expert on curves and level.


Here are some samples. The first is the HDR converted to 8 bits. The result is disappointing, but not horrendous. I have some "worse" examples that I will post ...when I find them.


Thanks for the input

[milt]

Robert,

Since I got this started, I will try to elaborate.  However, its clear to me that some of the other posters here probably in fact know more about HDR than I do, so I hope there will be other contributions.

I don't know for a fact just what formula any of the HDR programs uses to compute the HDR value for a pixel.  I've not seen any of them published.  I do know the formula differs from program to program.  For example, PhotoAcute claims that their formula is better than others and that you should use their program for "Step 1".  I can imagine some basic things about any such formula however.  First, you would have to decide on an exposure weighting to apply to each shot.  This can be done by looking at metadata, or it can also be done by looking the variance across the shots between the values for a single pixel and taking some sort of average of that variance.  For example, Photomatix used to use metadata, but now they seem to use the latter.  Then, when you have an exposure weighting for each shot, a nice simple and obvious formula would be to ignore all blown or null pixel values and just take a weighted average of the remaining values for each pixel.  I suspect in real life things aren't quite so simple, and that the improvements to this simple approach are proprietary to each HDR program.  One obvious complication is what to do with partially blown or partially null pixels, i.e. blown or null in some channels but not all.

Notice that there is a natural concept of "coverage" here.  You can say your bracketed shots have covered a scene if your shortest exposure shot has no blown pixels and if your longest exposure shot has no null pixels.  Thus there is at least one valid value for each pixel.  An observation made by some HDR practitioners is that even if you have covered a scene, you get a better result by doing so with more shots rather than fewer.  For example, even if -2, 0, +2 covers the scene, -2, -1, 0, +1, +2 will give you a better looking result.  What is probably going on here is a lowering of noise because of the averaging of more values.

Don't get frustrated over HDR if you are using CS4.  Nobody seems to think it is a serious tool.

--Milt--

[Steve Weldon]

Mode 2 is simpler, less powerful, and occasionally better. (I think this is what Steve calls "blending".)
--Milt--

I'd have sworn Photomatix called it blending.. so I brought up my version and it did.. but when I updated they now call it "fusion" which probably reflects the reality that the process while is essentially blending.. is indeed more complicated than that..

In any case.. I just gave "fusion" a quick go at it (sample below).. not bad, seems improved over blending.. but HDR gives you a great deal more control..

[Geoff Wittig]

Hi,

I have just started experimenting with HDR and am somewhat disappointed with the results so far.

Any advice out there? Am I doing something wrong?

PS:


Thanks

HDR requires lots of experimentation to get the look you want. The available software tools are all very idiosyncratic, with their own quirks and foibles. Photomatix has lots of sliders and apparent image controls, but IMHO the output always ends up looking very artificial and processed, almost regardless of the settings you choose. I've been especially disappointed with its handling of green foliage in the darker (usually foreground) parts of images, as it seems to make everything look like a plastic potted plant. With a lot of Photoshop massaging I can generally salvage something usable, but the aggravation is often not worth it.

FDR tools is worth a try. It permits you to choose which parts of your source capture are incorporated into the 32 bit image, and it seems to do a better job with natural color. Still looks a bit bizarre, though.

After just two days playing with it, to be honest, I'm much happier with Photoshop CS5's HDR pro feature. The results look a lot more realistic and natural to my eye using the default settings out of the box, and the adjustments are more intuitive. So I'll have to cough up the cash for the upgrade, darn it.

Finally, you may not need to use HDR at all. If the division between light and dark portions of the image are obvious and easily masked, a manual blend of exposures will work just fine. That way you're using all your favorite Photoshop tools for dealing with tonal range, rather than leaving it to some automated software "black box".

[feppe]

HDR requires lots of experimentation to get the look you want. The available software tools are all very idiosyncratic, with their own quirks and foibles. Photomatix has lots of sliders and apparent image controls, but IMHO the output always ends up looking very artificial and processed, almost regardless of the settings you choose. I've been especially disappointed with its handling of green foliage in the darker (usually foreground) parts of images, as it seems to make everything look like a plastic potted plant. With a lot of Photoshop massaging I can generally salvage something usable, but the aggravation is often not worth it.

FDR tools is worth a try. It permits you to choose which parts of your source capture are incorporated into the 32 bit image, and it seems to do a better job with natural color. Still looks a bit bizarre, though.

After just two days playing with it, to be honest, I'm much happier with Photoshop CS5's HDR pro feature. The results look a lot more realistic and natural to my eye using the default settings out of the box, and the adjustments are more intuitive. So I'll have to cough up the cash for the upgrade, darn it.

Finally, you may not need to use HDR at all. If the division between light and dark portions of the image are obvious and easily masked, a manual blend of exposures will work just fine. That way you're using all your favorite Photoshop tools for dealing with tonal range, rather than leaving it to some automated software "black box".

Excellent tips from Geoff, as always, and agree with everything he said, especially the last paragraph about doing your own blending.

I've also had good success with Tufuse Pro, which automates the blending without many of the complications true HDR brings. It doesn't seem to be supported anymore, though.

There is a beta program I tested earlier this week called Oloneo. Their disturbingly slick web page has some good sample videos. Just like many other HDR tools the UI can be overwhelming, but it seems to be geared towards realism rather than the cartoony HDR look which thankfully is giving away to the next fad.

[JohnBrew]

I have Photomatix. It's pretty good, but I found I got better results when using 3 bracketed shots instead of 5. Using 5 images gave me muddy shadows. Not saying that's the way to do it, but it's what worked for me with Photomatix. One other item I liked about Photomatix is that it takes RAW images and converts them at the same time (as you make HDR).

[BobFisher]

Sadly, you're working with your hands tied behind your back a bit with CS4.  Even the tools available in CS4 Extended aren't nearly as good as what Adobe came up with in CS5 HDR Pro.

I'll have a preview of Oloneo up on my blog tomorrow.  It's only in beta right now so not generally available anyway.

HDR Expose is another good one to have a look at if you're looking for a more 'natural' result.  It's the successor to the very good HDR PhotoStudio.  The backend processing has been updated quite a lot.  I think it's available for $100 right now.  Maybe that sale ended on Jul 31, I'm not sure.

Photomatix can give both natural and the hyper-real results but getting the natural result can take a little more work.  

HDR author Jack Howard did a two part video series for the Adorama Learning Center on soft tonemapping in CS4.  Since that's what you're using, it might be of some use to you.
Part 1.  Part 2.

[Robert Boire]

Finally, you may not need to use HDR at all. If the division between light and dark portions of the image are obvious and easily masked, a manual blend of exposures will work just fine. That way you're using all your favorite Photoshop tools for dealing with tonal range, rather than leaving it to some automated software "black box".

Well you may be right. I guess I was hoping there was a way of  having the same control and results as a manual blend with a lot less effort. Perhaps wishful thinking on my part. I will have to try some of the evaluation HRD applications and see what happens.

Does anybody have comments on the examples I posted - from a HDR point of view.

Thanks all for the info.

[BobFisher]

Sorry, I missed the examples earlier.  

What is it you're looking for?  The tonemapped result isn't that bad.  You might want to do some additional post-tonemap tweaking but what are you finding about it that's so offputting?

In looking at the source images, one thing I'd suggest is to bracket wider.  You've still got some dark shadows in your 'overexposed' shot and some too bright areas in your 'underexposed' shot.  The biggest key to good quality, natural results from HDR tonemapping is capturing enough brightness range.  This is the biggest problem most people run into and the biggest reason why they have problems getting a result they like.

The other thing to keep in mind is that the tonemapped image isn't necessarily the ending point.  It may be just the starting point.  Think of the tonemapped image as a new image that contains all of the range of brightness that existed in the scene when you shot it.  It may be a bit flat and lacking visual interest, but that's OK.  You now have the opportunity to create your contrast and create the visual interest as you saw it or as you want it.  The job isn't usually done with the tonemapping.  In many cases, the job is only beginning when the tonemapping is done.  Thinking of that tonemapped image as a new 'RAW' file might get you thinking about further editing and optimisation as you would with an actual RAW file out of the camera.

[EduPerez]

I toyed a bit with your images (hope you do not mind); just for comparison, "enfuse.jpg" is what enfuse yields using the default options, and "final.jpg" is what I get after some tone-mapping with pfstools.

[stamper]

I have just started looking at this process. I am not convinced that I will be doing a lot of it though. This link should be very helpful to you.

http://photo.net/learn/digital-photography...-photoshop-cs5/

In particular this passage sums it all up?

Quote

If you truly want to end up with realistic looking images—and are willing to invest the time this takes—it is best to take full control of the HDR blending process using Photoshop layers, blending modes, the Gradient Tool, and the Paint Brush Tool. This Hand HDR process is explained in Creating HDR Images by Hand and Creating HDR Images by Hand. Usually, Hand HDR means avoiding automated HDR tools such as Merge to HDR Pro—although automated HDR can be used to create a background layer, or to handle difficult areas like a forest with complex lights and darks.

Unquote

In essence HDR is only part of the process and not the start and the end of it as some examples on the web would have you believing.

[Bart_van_der_Wolf]

In essence HDR is only part of the process and not the start and the end of it as some examples on the web would have you believing.

Exactly!

Step 1 is the collection of the bracketed exposures.
Step 2 is the merging of those exposures into one HDR image. This includes (amongst others) ghost removal for subject motion.
Step 3 is tonemapping. This i a separate operation, also essental for LDR images IMHO.

Robert, from your 3 small JPEG exposure brackets, I merged and tonemapped the following example:
[attachment=23526:C0004_HDR_3_.jpg]

I used SNS-HDR Pro, a relatively new tonemapping application from a very talented Polish programmer. The program IMO still needs work in several areas, but it's tonemapping capabilities are the best I've seen sofar (and I've seen many).

Cheers,
Bart

P.S. at the time of posting the SNS-HDR website seems to be down, he is probably updating with a new version.