HDR tips?

[David Kilpatrick]

Olaf is talking about the proprietary Photoshop 32-bit High Dynamic Range image format, not the popular concept of the 'High Dynamic Range' picture - which he rightly identifies as being a wide dynamic range compressed into a normal one (8-bit depth, 16 million colours, 256 per channel etc).

The original Photoshop HDR file was 16-bits per channel, and you can open a raw file directly into this bitdepth (just select 16-bit when converting). This will use all of the raw data, especially if you do not apply a conversion curve of any kind (use Linear) and do not make ANY exposure, contrast, brightness, saturation etc adjustments. Within Photoshop, you can then manipulate all the available data since no raw format yet has more than 16 bits and most only use 12-bits, 3 channels.

After making the image look as you wish in 16-bit, converting to 8-bit does (in theory) the same as any two-exposure result can achieve. However, we all know from experience that creating the complex curves required to render the three-quarter to shadow tones differently from the mid to highlight tones is next to impossible. It's easier to blend two exposures.

To use Photoshop's current HDR - 32 bits per channel - two 16-bit conversions are needed, one of a dark image, one of light. Two 8-bit files are not enough.

I think the issue here is the difference between an extremely technical approach - creating a 32-bit file with a complete histogram even though no device known to man can reproduce this and the human eye can't view it - and making a dramatic pictorial image by combining highlight and shadow information which could not be captured simultaneously on film (or in a normal digital image).

I am no expert at all when it comes to either 16-bit or 32-bit HDR formats, as they have no bearing on my work, which is reproduced in print and would be perfectly OK with a mere 6 bits per CMYK channel like 1980s pre press systems!

David

[01af]

... you seem to say that it is not possible to produce an HDR image from any DSLR output, regardless of software, even if multiple exposures are taken?

No, that's not what I am saying.

Olaf is talking about the proprietary Photoshop 32-bit High Dynamic Range image format ...

Umm ... yes and no. I am talking about techniques to capture and store more than 16 bits per channel of dynamic input range. Of course, Adobe's HDR format is the only more-than-16-bit image format (up to 32 bit/channel) currently available to photographers that I am aware of. By the way, as far as I know it was originally developed to meet the requirements of the motion picture industry; it was not originally meant for photographers.

... not the popular concept of the 'High Dynamic Range' picture - which he rightly identifies as being a wide dynamic range compressed into a normal one (8-bit depth, 16 million colours, 256 per channel etc).

This concept is known to me as DRI (= dynamic-range increase). Of course, 'high dynamic range' and 'dynamic-range increase' eventually say the same thing only in slightly different words. But to me there is two different (albeit closely connected) concepts behind them ... or do you really use these two terms as synonyms? Maybe my understanding of the term 'HDR' is too Adobe-centered ... but in any case I feel it's essential to understand the conceptual difference between extra-wide-range image formats on the one hand and non-linear dynamic-range compression on the other, whatever the terminology is.

To use Photoshop's current HDR - 32 bits per channel - two 16-bit conversions are needed, one of a dark image, one of light. Two 8-bit files are not enough.

At least two 16-bit shots are needed indeed (with different exposures, not just different conversions from the same raw file), but more would be better. Similar to stitching several shots together for a panorama where generous spatial overlapping makes things easier for the stitching algorithm, some overlapping of the dynamic ranges will make it easier for the HDR algorithm to blend the single shots. So three or more differently exposed 16-bit images are better than just two.

You can also create 32-bit HDR images from 8-bit shots but then you'd need even more if you want to fully exploit the 32-bit potential---at least four, better six or seven.

I think the issue here is the difference between an extremely technical approach - creating a 32-bit file with a complete histogram even though no device known to man can reproduce this and the human eye can't view it - and making a dramatic pictorial image by combining highlight and shadow information which could not be captured simultaneously on film (or in a normal digital image).

Actually this is not two approaches. It's two completely different things. The latter (i. e. making a dramatic pictorial image) is the final goal in any case. In order to achieve this, you can go several ways. One possible way is the former, i. e. the HDR way. The 32-bit image here is just an intermediate---and grossly oversized---product on the way to a pleasing pictorial 8-bit image. Another way is blending layers manually---this is more tedious to do but will yield better results generally due to better tone control.

-- Olaf

[Beakydave]

... you seem to say that it is not possible to produce an HDR image from any DSLR output, regardless of software, even if multiple exposures are taken?

No, that's not what I am saying.

But you say no, or virtually no, digital camera is capable of 32-bit imaging. How then can one produce what you call a true 32-bit HDR from a camera that produces only 16-bit images? Or are you saying that if you combine 2 or more 16-bit files you can end up with full 32-bit image? I know that tweaking the 'exposure' of a single RAW is not the same as taking multiple bracketed exposures (but sometimes the latter just isn't posible). However, my limited use of CS2 HDR has given some very good results indeed from a single image even if, as you say, I have had to trick the software into thinking I have more than one exposure.

I have not witnessed any of the posterisation that you suggest, but perhaps that's because I have tended to use subtle adjustments only, just to bring the shadows up a bit and tone down skies - rather than going for the exaggerated effects that some prefer. In other words, some see HDR as a creative tool, I see it as a means of getting back to the true image.

I have used layers and masks as well, with equal success but to be honest, for my needs, the CS2 HDR route is much quicker. Yes, it's a bit of a blunt instrument but my HDR activity at the moment doesn't justify more sophisticated software.

Thanks for all your thoughts so far - it has helped my understanding a great deal - but I would like your comments on how to get a true 32-bit image from a 16-bit camera!

Dave

[01af]

But you say no [...] digital camera is capable of 32-bit imaging.

Not with a single shot. But with several, differently exposed shots of the same subject you can capture as many bits of dynamic range as you want.

How then can one produce what you call a true 32-bit HDR from a camera that produces only 16-bit images?

With a single shot, you cannot.

Or are you saying that if you combine two or more 16-bit files you can end up with full 32-bit image?

That's what I'm saying.

I know that tweaking the 'exposure' of a single RAW is not the same as taking multiple bracketed exposures ...

If you know that then why are you inquiring?

... (but sometimes the latter just isn't posible).

Sure ... for example when the subject is moving. Blending two different conversions from one single raw file will often yield just perfect results ... or at least good-enough results. But in this case, using Photoshop's HDR feature is a detour which does not really help achieving the best-possible image quality. However I do admit it's quicker and more comfortable to do than layers and layer masks (provided you don't mind playing tricks with the EXIF data). But it's not the right way.

-- Olaf

[Beakydave]

Thanks again, Olaf

It is all starting to make sense now. While I like to think I know how to take a good photograph, you will gather from my posts that the science is still a little beyond me. When I said that I knew taking multiple 'exposures' from a single RAW is not the same as real multiple exposures, that's all I was saying. I did not fully understand the implications in terms of bit depth - that is why I was enquiring.

If I read you correctly you are saying that multiple versions of the same RAW can only ever have the same 16-bit colours whereas multiple exposures will have different sets of 16-bit colours. In theory, then, 2 exposures with completely different 16-bit sets will merge to a 32-bit image. But I guess there will always be some overlap - if the two shots share the same 8-bit range, for instance, merging them will give a 24-bit image - right? Hence the need for at least 3, preferably more, exposures.

On one point I do disagree slightly though - you say that the method used by me is not the right one. As we all know, where manipulation is concerned, there are usually many ways of ending up with the same, or similar, results. Some are quicker, easier, may give better results, but who is to say what is the 'right' way of doing things? If the technique I use gives me exactly what I want, in the shortest time, then surely that is the 'right' way for me, even if others prefer another technique? It is little different from sharpening - I challenge anyone to tell me what the correct technique there is!

Again, thanks for your input.