Photoshop CS2 introduced the application’s first support for 32-bit high dynamic range (HDR) imaging. The support was pretty limited, consisting of the Merge to HDR command (for combining bracketed shots into a single image) and some basic imaging functions (cropping, cloning, conversing from 32 to 8 or 16 bits per channel). Even so, about a year ago examples started popping up of HDR experiments (not solely connected to Photoshop, of course, but helped along by CS2). In the time since then more good resources on the subject have emerged.
The Photoshop CS3 beta includes some improvements in the HDR realm. Some more functions (e.g. Levels) are enabled for 32-bit images, and the Merge to HDR command, although superficially similar to the one in CS2, contains a variety of improvements. It benefits from the new image alignment code; preserves a more complete set of source data; and uses improved algorithms for merging the data.
Trevor Morris has kindly supplied an HDR photo created with the CS3 beta, as well as the source frames. He says, "I could never get it to work in CS2, but it worked flawlessly in CS3, and I was quite pleased with the results." He writes,
This photo was shot inside the Christ Church Cathedral, located in Fredericton, New Brunswick, Canada.
For this particular shot, I used a tripod and remote to capture 12 exposures, from 1/125s to 20s, with a Nikon D70 @ f/16, ISO 200, FL 18mm. I combined the exposures using Merge to HDR, increased the local contrast, and gave the image a slight saturation boost.
Give it a whirl with your bracketed shots, and please let us know whether it works well for you.
I now tried HDR with auto-align in CS3 using two hand-held shots which differ greatly in exposure. And I have some bad beta news: CS3 did not manage to properly auto-align the two shots, they were badly misaligned — whereas the HDR function in CS2 did fine with these same images!
[Wow–that’s not good news. I’ll drop you a line to learn more about the images so that we can track down what’s going on. –J.]
So some more work seems needed here. But it’s promising to see that we now have Levels, Hue/Saturation and a few other editing options for 32-bit HDR images. So it’s coming along.
Maybe unrelated to hdr image quality, but do you have any idea if there is something we can do ourselves about jpg image quality and especially resizing quality ?
I know this issue is mentioned in the release notes, but these two things are keeping me from using cs3.
Save for web output is not as good as quality as cs2, and image resizing inside cs3 creates jagged edges, where cs2 are smooth and clean.
I’ve tried all settings, but cs2 produces better images every time.
[It’s a known bug & one we’ll fix, but unfortunately we don’t have plans to release an update to the beta. I wish it were much easier to issue updates than it is, but the process is quite laborious, and we need to focus on actually finishing CS3. –J.]
Apart from those issues cs3 looks fantastic (love the new icons too), I hated bridge cs2, and I love bridge cs3 (if bridge cs3 was released about a month ago, it would have kept me from buying aperture).
One small thing I don’t get is why all the applications need to have a completely different look and feel (photoshop – bridge – lightroom), what’s wrong with a single recognisable, clean, usable interface (photoshops new interface is very nice)
best regards,
peter
One point of clarity – I think it is somewhat confusing to present HDR images as “32 bit” in the same way that we think of 16 or 8-bit images in Photoshop.
This is because HDR image color data is stored rather differently, as 32-bit floating-point values which means a much larger range of potential values than would an image where color channels were held in straight integers as 16 and 8-bit images are.
[Yes, good point. I’ve linked to info on HDRI in the past, but it would be useful to post a little intro blurb for quick reference. Maybe I can wheedle that out of Chris, John P., or one of the other engineers sometime. –J.]
Also for those wishing to explore HDR, you may wish to look around at other standalone programs that can do a really good job of HDR assembly – Photomatix is I think one of the more popular commercial options, and there is an open source option called qtpfsgui that claims to be similar:
http://qtpfsgui.sourceforge.net/
Kendall – we tried a few ideas for how to label the new depth/mode. “Floating point” didn’t go over too well with most users, and “HDR” was an unknown concept to most users when CS2 shipped (heck, I spent weeks trying to explain it to my coworkers). But labelling it “32 bit” is correct, even if people think it’s limited to a [0…1] range. Then they discover that it’s not limited, and all sorts of magic starts to happen (and, unfortunately, some confusion). Even a lot of imaging experts took a while to grasp the possibilities of HDR imaging.
Now some users are familiar with the term HDR, but still not enough (and sadly, some of them think PhotoMatix is all there is to HDR). So, we’re really still getting started in this area.
We do document the fact that 32bit/channel is floating point in Photoshop: in the manual, online help, and SDK. (of course, nobody ever reads the documentation…)
I´m using Photoshop since CS2 to assemble my HDR images for the use in 3D. The results were as good and similar or better than with Photomatix or HDRShop, and I´m looking forward to have all (or most of) the PS tools available for my work. So please keep on
Trying the CS3 Beta to merge a n HDR panorama, the new automatic alignment led to a very bad result. There should be an option to disable this function.
Very good and necessary: The possibility, to load a preset response curve.
[Oliver, would you mind sharing sample files with us so that we can try them and determine why they’re not working well in CS3? –J.]
I’m not very impressed with the level of updates to HDR in CS3. Which is why i will continue to use Artizen. Photomatix is so over priced for what it does and HDRShop is to limited in it’s features.
Artizen HDR has full 8/16/32 bit editing and EVERY single function already works in all color depths such as Level, Color Balance, Layers, Brushes etc. etc…
So if Photoshop want to get serious about HDR they better pickup their game, not just updating an AutoAlignment function that still doesn’t work.
[How exactly does it not work? It’s far more useful to have some examples than to get what boils down to “you suck.” Otherwise there’s no way to validate your statement, or to make things better –J.]
Intriguing possibilities for HDR. Beyond just its use for pano or CG shots or those amenable to straight bracketing – the most dramatic and challenging results would seem possible during the magic hour. How then to mix parts of different sequences of bracketed shots, in such a rapidly changing light situation as this?
[Good question. It would be interesting to see how the new Auto-Blend functionality in CS3 might help here. –J.]
Interestingly for you all at Adobe (and does relate to the wider debate of HDR), is how the PS tools possibly address these semi/quasi/automatically is going to help decide for each photographer when he feels he’s crossed the line to “cheating” (think Heisenbug Principal).
[Heh–that’s a broad & interesting area. Some photographers debate whether it’s okay to clone out a soda can that appears in a frame, vs. removing it before taking the photo. Bring together two photographers and you’ll probably find three opinions… –J.]
Regarding the alignment problem, here´s the mentioned set of images for dl:
http://www.oliver-mann.de/projekte/cs3/Oliver_Mann_JPEG.zip
(They are reduced in size and depth, so I can´t guarantee that they behave the same way as my originals.)
I assume, the disalignment is caused by the slightly moving clouds. CS3 tries to align them, and this results in disalignment of the environment. Images taken from a tripod don´t need an alignment, and the algorithm seems to be very time consuming. Ideally it should be possible (not only for HDR) to define image areas that should be aligned. Also for speed and better control, it has to be possible to disable alignment.Another recommendation: Layers with layer masks would really make me happy 🙂
“I´m using Photoshop since CS2 to assemble my HDR images for the use in 3D. The results were as good and similar or better than with Photomatix or HDRShop”
You Obviously dont have a clue about Hdr Shop.
[I find that user perceptions of HDR software, processing, and output tend to be fairly subjective. Maybe that will change as the field matures. –J.]
The true potential in HDR imaging won’t be discovered until tools like Photoshop and Lightroom start using 32 bit float as the native processing resolution for ALL image processing operations.
Currently, any process has the potential to push numbers out of usable range or between fixed-point numbers, which means that you get detail loss with every operation you perform. Floating point precision goes a long way toward reducing that loss of headroom, and moving from 16 to 32 bits isn’t just a doubling of precision — it gives you exponentially more headroom.
In the digital music studio, the move from 16-bit fixed point audio to 32 bit floating point made a negligible difference in the perception of any given sound. Where we really noticed the change was in the final mixdown, after all the processing was applied. The files processed at 16-bit fixed sounded horrible compared to those processed at 32-bit float, because the 32-bit float precision suffered much less detail loss during processing.
Currently, I’m noticing a negligible difference between my HDR images and my correct exposures other than the way they’re tone-mapped. HDR captures more detail, sure, but it’s currently a pain just getting the HDR image to look as good as the correct exposure, let alone getting it to look better. I suspect there are a lot of newbies out there creating HDR images and assuming that they’re better just because they’re HDR, never thinking to take a detailed look at their correct exposures vs their HDR image.
When the native photoshop/lightroom processing format is HDR, I expect that to change dramatically, because we’ll be able to preserve the full range of HDR detail throughout the entire photo finishing process. I look forward to a future of the most stunningly detailed photographs ever created — but it’s not here, yet.
– Eric
Like Oliver I use 32bpc images for 3d, mainly as HDRi maps or reflection passes, but occasionally also for textures – I’d just like to re-iterate how joyous it would be to have layers and layer masks. In fact all tools would be nice, and also to be able to paint in values higher than 1.
[Say hello to Photoshop Extended. 🙂 This was something I couldn’t mention when I first posted this story, but PSCS3 Extended features full support for painting and layers in 32bpc documents. We think this makes sense in conjunction with the 3D, video, and technical focus of Extended. (We don’t expect photographers to need these features, as they’ll most likely tone-map down to 16bpc right after doing Merge to HDR. The difference is that photographers care about how the image looks on screen/paper, whereas FX folks care about feeding the extended data into other apps. –J.]
Re your reply to Jules:
[Say hello to Photoshop Extended. 🙂 This was something I couldn’t mention when I first posted this story, but PSCS3 Extended features full support for painting and layers in 32bpc documents. We think this makes sense in conjunction with the 3D, video, and technical focus of Extended. (We don’t expect photographers to need these features, as they’ll most likely tone-map down to 16bpc right after doing Merge to HDR. The difference is that photographers care about how the image looks on screen/paper, whereas FX folks care about feeding the extended data into other apps. –J.]
Hmmmmm…..
Not sure I agree that photographers don’t care about feeding “extended” data to other apps – PRINTER DRIVERS, for instance. I’m running Canon’s iPF5000, which has a 16bpc input plugin. I can see a day approaching when 32bpc print input will be available and I would really like to have my HDR photo data start and stay in 32bpc mode. Having a full set of edit tools that work on 32bpc files is really attractive. I am considering shelling out for PSCS3 Extended, but will do some more research first (if I was wealthy it would be a done deal).
John 2 questions. Why when I process an HDR does it come up as sRGB even when my Camera Raw default is ProPhoto? And why are the images processed in CS3 so much less saturated than images processed in CS2? I have had to add the step of Processing into 32 bit then adding 50+ saturation and then converting to 16 bit to get them close to what they were in CS2.
And for those that want to turn off the auto align in HDR just open a bunch of images in PS then do Automate>HDR and turn off the alignment there. Then when you go back to Bridge to run HDR it will be turned off.
[Lewis, I’ll check into what you mention & try to post more info here. I haven’t heard of problems with or changes to saturation in CS3 HDR. –J.]
John it is strange, I can process the same files in CS2 and CS3 doing the same procedure and I get very different results. I don’t remember that being the case with the public beta but I am using a released version (golden master) now.
I’m a photographer and I’d dearly love to work in 32-bit natively. Frequently I clip an image’s histogram using an adjustment, and once I have there’s no way of retrieving the detail without creating another copy of the original image and processing it differently. A very common scenario is a sunset. The ground is dark, the sky is bright. The detail is there in the raw file but there’s no way to load it all in one go. To get the detail I have to open the image twice using different settings, then merge them together afterwards. It would be great if I could open it once and edit the image selectively instead.
[I think what you’re looking for is part of Photoshop Extended. Our expectation is that photographers typically don’t need or want to remain in 32-bit mode; rather, they want to bring in the full range of data, then tone-map it down to something that can be printed and seen on screen at once (8 or 16-bit). Film and 3D folks, on the other hand, are the ones who want 32-bit input/output, which is why that feature is part of Extended. –J.]
I find that I don’t get enough control with tone-mapping. I get better results using adjustment layers and masks to achieve selective contrast enhancements.
Hmmm…I just tried the paint operator in Toxik and as far as I can tell it behaves about as oddly as Photoshop when it comes to floating point colors.
I imported an image in Toxik and hooked it up to the paint operator. I selected the color (0.99999, 0.99999, 0.99999) using the color picker which admittedly LOOKS much more professional than Photoshop’s. I start to paint. Then I go back to the color picker and sample the color of the stuff I just painted. WHAM! It says (0.99950, 0.99950, 0.99950) all of a sudden!