Understanding native tonal response

Sunrise over Lake Michigan

Continuing this little series on tonality, mood and monochrome, I’d like to explain a little about the idea of native tonal response: it’s something I’ve frequently referred to in reviews, but never fully explained. Unfortunately, there are a very large number of variables, so bear with me.

Tree exhalation

We first need to take a step backwards to understand how digital sensors work. Light excites a photosensitive layer into releasing an electron (or more), which is then released once an electrical ‘gate’ (the transistor) is opened, allowing a current to flow. This current is measured and then translated into a value. That particular location has a physical x-y address, and so do its neighbours; to complicate things further, each adjacent physical address has a different color filter over the top of it, to allow measurement of the intensity of light of a different wavelength. With the exception of Foveon sensors, none of these devices can measure the wavelength and intensity of light at a single physical location accurately; interpolation from adjacent locations is required to come up with an approximate value for the color at that location. Our displays use the R-G-B primary colours in various combinations to make up every intermediate color (or close to it); the sensors therefore also take readings and output R-G-B values for a given physical location. To complicate things further, this interpolation may be done in computer or in camera, and by different software. And then at the output end, we could be viewing the final image in print, on screen, at different sizes, and at different levels of information compression.

So what actually affects tonal response, and more importantly, how much control over the process do we really have?

Evening ruins

The idea of tonal response can be distilled down to the following concept: for a given absolute amount of light hitting the sensor, what value does the sensor record? It is an input-output map, nothing more. But there are so many things affecting that input-output map that no one single element really dominates. Firstly, there’s the sensor type: CCD, or CMOS? CCDs tend to be nonlinear in response; nothing much down the bottom, as though the activation threshold is high and the low-light read noise (i.e. erroneous/ spurious signal) is prevalent, but then tonal response appears more natural towards higher luminance levels. CMOS sensors are more linear in their tonal response, which is better, but this also means visually more abrupt clipping at highlight and shadow boundaries (which is worse). And then you’ve got the color pattern itself – multilayer Foveons respond differently to Bayer patterns, and then there are three (RGB) and four (RGBE) color Bayer patterns, and other pattern layouts – X-Trans, for instance. And we haven’t even talked about the read or on-chip noise reduction circuitry between photosite and raw signal; on-camera image processing to raw or JPEG, your choice of raw converter, and postprocessing.

The call

It is actually quite common for the same base sensor to have very different effective tonal response depending on the camera it is put into: this is the effect of the processing engine, plus how your choice of raw software or postprocessing affects the final output. Even if you zero out everything in a universal raw converter like ACR, you’ll probably find two cameras look very different under the same circumstances – this is partially to do with whatever defaults ACR has applied at the behest of the manufacturer or themselves; or it may be to do with the in-camera processing engines. And I’m sure we’re all very aware that not all processing engines are equal. ACR, for instance, has become very proficient indeed with anything Bayer; it’s useless on Fuji X-Trans files before version 8 and doesn’t even bother with Foveon.

Other third party software may do a better job with the actual interpolation from raw data to a viewable image, but they do not offer the same workflow as ACR: for instance, it is very difficult to apply local or other adjustments without outputting an image first, which means that those adjustments happen on data that has already been worked on; it isn’t as ‘close to the source’ as ACR might be. This may not be visible under most circumstances, but for processing that’s close to the limits – such as extreme contrast changes, borderline out of gamut colors, or very fine tonal transitions in deep shadows or bright highlights – that slight loss of data may well become visible and turn into something make or break: think posterization.

Black Island

A couple of examples of cameras with different ‘tuning’ but the same, or very similar sensors would be the Nikon Coolpix A and Ricoh GR, the Nikon D800E and D810, and the Phase One IQ250 and Pentax 645Z; I’m sure there are plenty of other examples but these are the ones I’m most familiar with. The IQ250 has a very different color palette to the 645Z, but these could easily be fixed with a little profiling; interestingly even though both cameras lack an AA filter, they produce results of very different sharpness – my theory is that there’s a bit of unsharp masking or something similar going on before the raw files are written. There isn’t any more resolution, because we have the same spatial frequency of sensor, and we were using the same lens via adaptors on both. The Pentax will take a lot more sharpening before hitting the limit; the Phase One, less – and sharpened optimally, they both look identical.

Moon and bridge

The D800E and D810 are slightly trickier: it isn’t clear if they’re using the same sensor or not, but it’s highly unlikely the D810 uses a new design as we’re not really seeing any improvements in color accuracy or noise performance. But what is telling is we moved a generation of image processor between the two cameras, and suddenly we’ve gone from a very linear sensor in the D800E to a very nonlinear one in the D810. The D800E favours underexposure to maximise dynamic range: you get perhaps a stop of recoverability at the highlight end, but seemingly limitless shadows that not only retain color information and relative tonal separation/contrast when brightened, but also very low noise. The D810’s shadows look muddy and flat when treated the same way, but what you think is clipped at the highlight end isn’t – in fact, you could probably have pushed another stop or stop and a half. (We are of course testing with the same raw converter, settings and lenses here.)

In practical terms, this means that the D800E looks better in low light or high contrast because you can pull out hints of detail here and there with little noise penalty. The D810 is a bright-light camera – highlights roll off very naturally, but maintain relative tonal separation well into the higher zones where the D800E would just have clipped. It means in practice that whilst I’d never really liked the D800E for monochrome work – but it was manageable because of its enormous starting dynamic range – I find the D810 a very natural camera for B&W work, simply because the highlight rolloff is so much more gradual. I don’t need to do anywhere near as much dodging and burning, and a single curve will suffice where I’d had to use two or more, previously. There are reasons and jobs that justify me keeping both.

Departure time

I’ve left the most stark example for last: the Coolpix A and GR. Both cameras definitely use the same 16MP APS-C Sony sensor, but one has been tuned for color, and one for black and white, and this obviously shows. Once again, to understand why, we need to go back to human psychology and physiology. Different colours appear to our eyes of different relative brightness even if they are of the same absolute luminance level; this is mainly to do with the way our retinas respond to the different wavelengths. There will naturally be attenuation towards the ends of the visible spectrum, which is something that digital cameras do not suffer from – in fact, it’s just as easy for them to pick up UV and IR pollution as actual intentional signal. This is one of the reasons that cameras have such strong UV and IR cut filters over the sensors; otherwise, the output would not look natural at all. You’ll actually find that a lot of color improvement in digital cameras has come from these cut filters both increasing in strength and ability to approximate the response of our eyes.

Lonely chef

Cut filters aside, this means that a camera tuned for color is going to have uneven amplification for each channel (this happens somewhere between the raw data and writing a raw file, in the camera’s on-board converter) in order to deliver a natural looking image. This can distort the relative luminance differences between areas and produce a file that’s very flat and lacking in local contrast; i.e. poor for B&W conversion. However, a camera that has equal amplification across all channels is going to be great for B&W but produce very strange and possibly undersaturated-looking color – and that’s precisely how the GR’s color images look without profiling. For cameras with larger photosites and more dynamic range, such a choice isn’t necessary: it can be profiled out because you have enough information to work with.

Texture medley

A true monochrome sensor will get around these limitations – but again, there’s the question of the output map, and you cannot have color if you change your mind for one shot. The upside is that you can of course optimise that output map for the specific kind of nonlinearity that looks good for monochrome images.

How much of the difference can we make up in postprocessing? To a large extent, that depends on the amount of information your sensor is capturing. The more dynamic range, the easier it is to reallocate the tonal response the way you want it. It’s one of the reasons why you never hear about MF sensors being poor at one or the other – it’s just not possible to make a poor B&W file if you’ve got nearly fifteen stops of dynamic range. It might need a little bit more work to introduce a similar nonlinearlity to match our human visual range, but the data is there. If the data isn’t there, no amount of tonal stretching through curves and dodging and burning is going to avoid posterisation or put in subtle color gradation where there is no information to begin with.

Father and son at the end of the day

Color response is fortunately much simpler to fix: it can be profiled out, and ideally done for different ambient color temperatures (again due to nonlinearities in amplification or differences in the Bayer filter itself). Bottom line: you can probably get a 95% result with a non-ideal camera, but it will take you many, many times longer than a 100% result with a fit-fo-purpose one. My advice is to think very carefully about the types of subjects you shoot, and the light conditions – and select equipment accordingly. The images in this article have been specifically chosen to demonstrate this: I was able to easily (with only a minute or two worth of PS work) achieve the color or tonality (and more importantly, color and tonal separation) I wanted. MT

Color profiling for cameras is covered in Photoshop Workflow II, and tonality and style are explored in Outstanding Images 5: Processing for Style. Both are available here from the Teaching Store.

__________________

Take your photography to the next level: 2015 Masterclasses now open for booking in Prague (9-14 Mar 2015) and Lucerne (17-22 Mar 2015)

__________________

Limited edition Ultraprints of these images and others are available from mingthein.gallery

__________________

Visit the Teaching Store to up your photographic game – including workshop and Photoshop Workflow videos and the customized Email School of Photography; or go mobile with the Photography Compendium for iPad. You can also get your gear from B&H and Amazon. Prices are the same as normal, however a small portion of your purchase value is referred back to me. Thanks!

Don’t forget to like us on Facebook and join the reader Flickr group!

Images and content copyright Ming Thein | mingthein.com 2012 onwards. All rights reserved