Tone mapping?

Ok, I'm done hearing this term and not understanding exactly what it is...can anyone explain this term in detail, how it affects a photo, when does it matter, implications for workflow, when to think about it, etc. Thanks.

It is a term used in HDR --High Dynamic Range-- processing. An HDR photograph is an image created from several photos with bracketed exposures (under + over exposed) to capture a large dynamic range in a high contrast scene. The images are combined with HDR software (CS3 and Photomatix are two examples). However, the HDR software cannot properly display the image until it is tonemapped. Tonemapping means adjusting the HDR image so it is pleasing to the eye and doesn't look overdone. It also means converting the image to 16 bit so all the tones can be seen on the monitor.

Alain Briot

Alain,

It also means converting the image to 16 bit so all the tones can be seen on the monitor.

Click to expand...

This last sentencence is not correct. Monitors are not able to work in 16 bit mode, but in 8 bit mode.

Tonemapping is = compress HDR image into LDR (Low Dynamic Range) image, preserving the general apparence of the HDR.

Alain Briot said:

It is a term used in HDR --High Dynamic Range-- processing.

Click to expand...

Yes, although it is not restricted to that.

In general terms, one can look at an image as a recording of a range of luminances that existed in the original scene. Some scenes have low contrast (subject contrast multiplied by lighting contrast), while others span a huge range of different luminances.

Ideally, for a realistic rendering of the scene, one would like to recreate the same range of luminances when we look at the resulting images, be it on screen or in print. However, since the display modalities do not allow to produce arbitrary amounts of luminance, we'll have to compromise. We'll have to apply a tonemapping function to translate from input luminances to output luminances.

Printed output is typically limited to allow an up to 100:1 ratio, displays on average range from say 80:1 to 500:1 (sometimes even higher).

So what to do when we have an original scene with a contrast range of say 50:1, and we want to print it on a glossy paper with a 100:1 capability? Well, to reproduce the original contrast sensation, we can print it at the original scene's contrast, but we can print it a bit darker or lighter. It will still fit in the capabilities of the medium but the mood changes. We can also choose to transform the original scene to a more contrasty one, not as realistic anymore but with more punch. We then stretch the range of luminances to cover the full capability of the paper. This is typically done by adjusting the black point and white point (with a levels adjustment on the histogram).

We can even choose to increase the contrast for some tones, and/or reduce the contrast for others, or to brighten some tones and darken others. This is typically done with a curves adjustment or burning and dodging techniques. While this will violate a realistic reproduction, it does allow to emphasize certain structures in a more creative fashion.

Now, what to do when the original scene exceeds the capabilities of the paper. First of all, we may have a problem with recording such a contrast range. Technically, most DSLRs are limited to recording 4095: ratios, some can theoretically reach a 65535:1 ratio. Seems enough one would guess, but there are many scenes that exceed such a range, and shadows in such large ranges have a lot of noise, so one tends not to push the envelope to the maximum capabilities. The HDR techniques specifically address these issues and can record much higher ratios.

Let's assume we have a scene with a 2000:1 contrast (e.g. fill flash was used to improve the shadows). The straight forward way of dealing with it, and be able to print it, is to scale down the contrast to fit the 100:1 paper capabilities. Technically not difficult, with a linear tonecurve and a Levels adjustment, but it will result in a low contrast and dull looking image.

This is the most common scenario (HDR is an extreme version of it) requiring a non-linear tonemapping function to kind of shoehorn the large range into what we can achieve in output, while still maintaining the look of the original scene. This is also an interesting field in imaging research, and it allows a lot of creativity.

The tonemapping function that allows to maintain the original contrast sensation is scene dependent, highly non-linear (it rather is adaptive to local contrast throughout the image), and also has to do with the human visual system and perception. And then there is also the creative aspect where one deliberately wants to deviate from a perceptually realistic rendering, e.g. to emphasize certain aspects of the image.

So we end up with tools and techniques that allow us to map the different tones in the original scene to tones in the output (AKA tonemapping), sometimes be cause we have to, but hopefully more often because we want to improve the output quality. This also implies that one could/should adapt the transfer function to the output modality's capabilities, and the viewing conditions.

Bart

Can anyone give a real life example of using tone mapping? I've also heard this tone-mapping term for years and people use in so many different ways I'm thinking it either means many things or poeple use the term incorrectely.

I'm pretty dense on this one perhaps...maybe seeing a reallike example of tone-mapping in action would reduce the many vague aspects.

At the following link, you can see an example.

http://www.photoresampling.com/hdr_eng.php

OK, thanks...so tone mapping is HDR. I've heard this term used in many other instances not talking about HDR so they must be incorrect.

Hi Shane,

so tone mapping is HDR.

Click to expand...

Not necessarily. Read the first sentence of Bart's post. tone mapping is used to reduce an HDR image to LDR which can be displayed on the screen.

Here's a couple of screen shots showing Photomatix in use.

The first one shows four exposures stacked behind the resulting 32 bit per channel HDR. Strange colours! That's because the monitor can't possibly display a 32-bit image.



The second one shows the result of applying tone mapping to the HDR.



Regards,

Stuart

Hi Shane,

You've drawn the wrong conclusions.

Tone mapping is used by HDR, but tone mapping can also be used elsewhere. (a horse is a four legged animal, but not all four legged...etc., and others)

Best wishes,

Ray

OK, am using the now old friend Google...found this through Wikipedi...http://www.cs.utah.edu/~reinhard/cdrom/

Guess I've got some reading to do.

Would it be correct to say that most tone mapping today is done through HDR to capture proper exposures for different tones in a single image?

Basically, using a curve in PS is tonemapping already.

Photomatix has a PS-plugin - for 16bits - also; it's called tonemapping.

Shane Canfield said:

OK, am using the now old friend Google...found this through Wikipedi...http://www.cs.utah.edu/~reinhard/cdrom/

Would it be correct to say that most tone mapping today is done through HDR to capture proper exposures for different tones in a single image?

Click to expand...

Yes. Tonemapping and HDR are used interchangeably by many people. I like to think of tonemapping as being the "mapping of the tones" of an image to a reduced contrast range that will display properly on the monitor and print nicely.

WHat's interesting with HDR is that the same HDR image can be tonemapped in a variety of ways, to the point where there can be large differences between the different version of the same image depending on how the tonemapping was done.

For me one of the main issues of HDR is controlling the tonemapping process. The more control we have, the better the results become. The problem with early HDR applications was that we had little control over tonemapping. The latest versions are much better, although we can go way further.

How about layer adjustments in the HDR applications to allow for local contrast control for example? Right now we have to do that in Photoshop, after completing the tonemapping.

Another tonemapping application you can look into is Lightzone. It actually offers layer masks and local selections. Lightzone isn't a pure HDR app in the sense that it doesn't let you combine several photographs, but it can "rescue" a very high contrast image in very powerful ways. It's also a raw converter so you can go from raw to tonemapping in the same app then polish the results in Photoshop.

What you'll find out with HDR is that you need a raw converter, the HDR app and photoshop to do a professional job. It's not a single-app process, at least not yet.

Alain Briot

Giovanni Brembati said:

Alain,

Tonemapping is = compress HDR image into LDR (Low Dynamic Range) image, preserving the general apparence of the HDR.

Click to expand...

Thank you. That's a good way to put it.

Giovanni Brembati said:

This last sentencence is not correct. Monitors are not able to work in 16 bit mode, but in 8 bit mode.

Click to expand...

I agree, but monitors can display 16bit images properly even though not all the colors are visible, while they cannot display 32 bit images properly at all. That's what I was getting at.

ALain Briot

Hi, Shane,

Shane Canfield said:

Ok, I'm done hearing this term and not understanding exactly what it is...can anyone explain this term in detail, how it affects a photo, when does it matter, implications for workflow, when to think about it, etc. Thanks.

Click to expand...

Conceptually, it means modifying an image so that the different "tones" (and "tone" is an ambiguous term, but perhaps we can think of it as "the luminance aspects of the colors") that can exist within the range found on the original image are each "mapped" onto a (generally-different) tone in the modified image, usually following some smooth "curve" that defines the "mapping".

Thus, the term is actually applicable whenever we change the "brightness", "contrast", or "gamma" of an image.

However, the term is most commonly applied to the process when applied, during postprocessing, to reduce the span of "tone" found in an individual original image to fit within the gamut of some destination output medium, or to exploit that gamut to the desired artistic effect.

I agree, but monitors can display 16bit images properly even though not all the colors are visible, while they cannot display 32 bit images properly at all. That's what I was getting at

Click to expand...

Alain,
the problem more than bit depth is contrast ratio.
Camera raw can capture 12/14 bit of data and if the scene dynamic range is not very high, you can develop more image with different exposures and create HDR from them.

In any case, monitors are not able to display 16 bit images. To display a 16 bit image you must "tonemap" it. The more simple way is rescaling, but rescaling is a tonemapping.

Radiances recovery (the first step for HDR) is performed in floating point (32/64 bit per channel).

What you'll find out with HDR is that you need a raw converter, the HDR app and photoshop to do a professional job.

Click to expand...

I disagree.
Shoting in raw mode is not required. Jpg multiple exposures may be used.

Giovanni Brembati said:

I disagree.
Shoting in raw mode is not required. Jpg multiple exposures may be used.

Click to expand...

Well, true HDR is 3 x 32-bit - based on 16 bit images. Still, camera- jpgs, converted to 16 bit can be used, but 16bit tiffs make real sense.

Michael,
as I said, HDR may be 32 or 64 bit per channel, but the photos you use for creating hdr can be 8 bit instead of 16. Tiff is not a special format, you can use any format you like, jpg too.
In Debevec and Malik work (SIGGRAPH 97) images are 8 bit per channel.

I've actually wondered about the relative quality of HDR processing done from jpegs, tiffs or RAW files. So I made comparison tests with the same images converted to Jpeg and to tiff and with doing HDR directly from the Raw files, converting the RAWs in the HDR software (Photomatix in that instance).

The differences in quality are noticeable, and the finest quality is obtained by doing HDR straight from the Raw file, converting the Raws in the HDR software. Why? Because the software has more information to work with and because it is working with single-channel files instead of tri-channel files. The most noticeable difference is that more details are visible in the brightest highlights and shadows. This was with a very high contrast scene, but then again I currently only use HDR for extreme contrast scenes whose dynamic range cannot be captured with a single exposure.

This matters to me because my goal is the finest image quality I can create. It is important for fine art. It may not be as important if the goal is not fine art or if quality is not the primary concern. It all depends on what the goal is.

HDR raw conversions

I have also found that in many instances I will get a better result by allowing the software to make the raw conversion, especially in photomatrix then detail mapping to get the final result. The tone mapping in photomatrix doesn't seem to give me the same amount of range. However it's also very image dependent. Many times will detail and tone map to see which version looks better.

BTW does anyone know if the photomatrix beta 4 will open 1ds MKIII raw? I was able to borrow a camera last weekend and have some raws I wanted to work up. Photomatrix's current official version will not open them.

Paul Caldwell

You may get several examples on this already. For good examples/gallery - go to www.hdrlabs.com - and check out Mr. Blochi's his book on the subject.

For real life example - see this demo: http://dativ.at/fotos/panoramas/adr/adr.html

I know you may say that this is an ADR etc...and not HDR...however, the concept is the same.