Film color balance

[Tim Gray]

This is more of a technology question. So I understand that if you use daylight balanced film under tungsten light, the colors are off. Furthermore, it's impossible to correct fully in post processing, either with filters in a darkroom or corrections in Photoshop. What makes correcting white balance on digital raw files superior and more correct?

As I understand, digital sensors (for the most part) have RGB filters in front of individual pixels, and the image is reconstructed from the Bayer filtered data. If I got it correctly, adjusting the white balance on a raw file is essentially adjusting the gain on the individual R, G, or B channels *in post* to correct for color temperature differences. Why can't you do this the same way in Photoshop (or with filters in the darkroom) with a color negative. Either way, film or digital, you're capturing the scene light, with it's specific color temperature, on sensitive media with similar RGB filtering (at least spectrally) and set gains/sensitivities on those channels at capture time.

 

[PHOTOTONE]

The color dyes in color film overlap each other (one on top of another) rather than discrete pixels like digital. If you correct for one color balance error you affect the reproduction of the other two dyes. You can balance color film shot under wrong lighting for pleasing skin-tones (for example) but you will find that not all colors are corrected the same amount, hence while the skin-tones will look good you will probably have an off-color look to neutrals, shadows or other parts of the image. Some of this effect is called "color crossover".

In digital the primary colors are discreetly captured by individual pixels, thus when you increase the gain on say the RED channel, you are not affecting the other two channels. When you scan film, while the capture is by discreet channels the original film is not, the three color dyes are one on top of the other, and this limits the amount of correction you can do.

 

[PHOTOTONE]

Its like a see-saw. If you shoot color film under a light it was not factory balanced for: In printing or scanning you increasing the strength of one color (from a color film image) you are suppressing the other two color records. If you achieve a correct balance for your mid-tones, the shadows and highlights will be out-of-balance.

By shooting the "correct" film for the lighting, you render the three color dyes in close enough balance that correction for optimum balance is possible.

You can pre-filter your exposures on the camera by using the color balancing filters designed to allow shooting tungsten film under daylight, or daylight film under tungsten. This has been done for decades in motion picture work, where the film stock is always factory balanced for tungsten, and it is always shot outdoors with a balancing filter. This allows one film stock to be used for a whole production, both location work and studio shots.

 

[Tim Gray]

Ok, that kind of makes sense. But in photoshop for example, you can just change the red channel's gain. So, I guess the red sensitive layer in film is not just sensitive to red. More to the point, is producing an accurate red for example with film dependent some what on the blue and green sensitive layers as well? Whereas on digital, red is solely the domain of the R pixels? Put another way, do the dye layers in negative film interact together to produce accurate color, where digital is more like taking three separate exposures, one for R, G, and B?

 

[Photo Engineer]

You can never fully correct an improperly exposed film in terms of color balance. This is a matter of the mismatch of the color curves which can never be properly realigned. A full explanation with examples is given in the book "Post Exposure" by Ctein.

Digital is just a poor mans fix which does not always work.

PE

 

[PHOTOTONE]

Ok, that kind of makes sense. But in photoshop for example, you can just change the red channel's gain. So, I guess the red sensitive layer in film is not just sensitive to red. More to the point, is producing an accurate red for example with film dependent some what on the blue and green sensitive layers as well? Whereas on digital, red is solely the domain of the R pixels? Put another way, do the dye layers in negative film interact together to produce accurate color, where digital is more like taking three separate exposures, one for R, G, and B?

Yes, the dyes interact to produce the colors, as they are layered one on top the other, they are not discrete. In digital individual pixels have individual color filters over them in a pattern (a Bayer pattern) and the pixels behind the RED filters (for example) just record the red content of the scene. Thus when you adjust the RED channel, you are not affecting the other two color records.

In a color print, for example it take two dyes to make red: Magenta and Yellow.

 

[Tim Gray]

Ok I think I got it now. I had this mental picture of color film working like three independent exposures of R,G,B filtered B&W and layered, but done so conveniently in one film. But the dyes interact to make the colors.

Its a shame that fast tungsten balanced film isn't made anymore. I wish something akin to Kodak's Vision3 500T was available in 35mm still.

 

[Photo Engineer]

Well, what actually takes place is that digital systems have a white balance and internal software that fix the curves on the fly in-camera to compensate for exposure errors such as color temperature. Film is a fixed recording medium but digital is dynamic. Therefore, in film, you have a fixed characteristic or response curve and in digital, you do not.

PE

 

[Tim Gray]

Well in digital, you can compensate for color balance in post. In the raw file, the data is the data - just the R,G, or B value at that pixel. Each pixel + filter must have some set spectral sensitivity and overall gain. While you can adjust in camera on the fly, you can also do it in post with the raw file. So it might not just be adjusting the R, G, and B gains, but also the contrast curves by the appropriate amount, but it is and can be done in post. And most often is done that way.

Can you properly correct film color balance in post if you correct both the relative levels and the contrast curves?

 

[tim_walls]

Well, what actually takes place is that digital systems have a white balance and internal software that fix the curves on the fly in-camera to compensate for exposure errors such as color temperature. Film is a fixed recording medium but digital is dynamic. Therefore, in film, you have a fixed characteristic or response curve and in digital, you do not.

Thinking out loud, I would speculate that this colour balance compensation/adjustment in a digital camera takes place in the digital domain - it is not done through any analogue amplifier adjustment at the time the photo is taken.*

Taking that as a given, I would presume that this colour balance adjustment is not 'free' - i.e. the process of adjusting the curves must be a lossy process; some additional errors must be introduced when the already quantised data is then adjusted.

Which would tend to suggest a properly exposed film (i.e. correct film used for correct job) should give better colour fidelity than a digital camera unless you're fortunate enough to shoot the digital image at the 'native' colour balance of the camera (if there even is such a thing, and if you knew what it was.)



* I posit this on the basis that all the digital cameras I've ever used have output as their 'RAW' file an uncorrected image, with the white-balance correction being done at the RAW file processing stage. It is just speculation though...

 

[Tim Gray]

Tim, I agree with you. In fact I've found some articles over on the libraw.org site talking about this problem. They have an article about shooting with the appropriate magenta filter to basically set the gains on each channel to 1. They also talk about how the noise in certain channels in certain color balances can contaminate the whole picture. For example, shooting in tungsten light, the blue channel is going to be underexposed, so when you look at the non-white balanced image, the blue channel is noisy. Correcting the balance distributes the noise somewhat to the other channels.

 

[Photo Engineer]

Thinking out loud, I would speculate that this colour balance compensation/adjustment in a digital camera takes place in the digital domain - it is not done through any analogue amplifier adjustment at the time the photo is taken.*

Taking that as a given, I would presume that this colour balance adjustment is not 'free' - i.e. the process of adjusting the curves must be a lossy process; some additional errors must be introduced when the already quantised data is then adjusted.

Which would tend to suggest a properly exposed film (i.e. correct film used for correct job) should give better colour fidelity than a digital camera unless you're fortunate enough to shoot the digital image at the 'native' colour balance of the camera (if there even is such a thing, and if you knew what it was.)



* I posit this on the basis that all the digital cameras I've ever used have output as their 'RAW' file an uncorrected image, with the white-balance correction being done at the RAW file processing stage. It is just speculation though...

Tim;

I believe that all of what you say is true, but I'm not sure about it being lossy. It is lossy in film due to the fact that the toe and shoulder portions of any one layer cannot be moved horizontally, but in digital it can and this can reduce the loss.

If a film layer shoulders out or toes out in the middle of the curves of the other two, you cannot fix it even with Photo Shop or any other digital means, as the data is lost due to the flatness of the single layer's curve at that point.

PE

 

[Tim Gray]

So white balance works better in digital because it's a linear capture medium, unlike film? Linear as in having no toe or shoulder.

I know you can white balance 3 chip (or 4:4:4 color sampling) digital as well, so its not because each pixel is only capturing one color.

 

[Photo Engineer]

So white balance works better in digital because it's a linear capture medium, unlike film? Linear as in having no toe or shoulder.

I know you can white balance 3 chip (or 4:4:4 color sampling) digital as well, so its not because each pixel is only capturing one color.

Tim;

Digital has a toe and shoulder which is plotted as V Log E (Voltage vs Log Exposure) as opposed to D Log E for analog where D = Density.

The difference is that the digital scales can be moved electronically back and forth to some degree and this is often done in-camera, but cannot be done in film. Digital is like having 3 separate images on 3 sheets of paper. You can move them back and forth to a certain extent. But, on the horizontal portions of the curves, data is still lost in both analog and digital.

Get the book by Ctein. It gives very good examples.

PE

 

[Tim Gray]

I ordered it about 2 hours ago

 

[tim_walls]

Tim;

I believe that all of what you say is true, but I'm not sure about it being lossy. It is lossy in film due to the fact that the toe and shoulder portions of any one layer cannot be moved horizontally, but in digital it can and this can reduce the loss.

Moving the curve shouldn't be 'lossy' with digital - if you ignore clipping at either end. 'Stretching' the curve wouldn't be lossy per-se, but since you're starting out with integer data, you will introduce banding artifacts as you create gaps in the data. Compressing the curve at any point on the other hand will be genuinely lossy, insofar as once you've done it you can't then reverse the process to get back the original gradation.

At least, if I'm thinking about this right... Which I may well not be .


Part of the 'solution' to this in digital terms is of course the one taken by Apple's Aperture or Adobe's Lightroom, which is to store the adjustments as a series of steps to be followed (i.e. transformations to apply) rather than 'fixing' the image after each step. I.e. if the colour balance correction is only one adjustment of many, you can process all of the adjustments in floating-point arithmetic and then only 'fix' the image back to quantised integer values after the last step is applied; this eliminates the accumulation of errors as successive adjustments are made, but doesn't eliminate the fact that if you start with quantised data and then make these sorts of adjustments you will introduce artifacts of one form or another.

Of course the other simpler part of the solution is to sample the original data at a much higher resolution than your intended output. I have a feeling there's probably an application of Nyquist/Shannon theorem in here somewhere but it's gone midnight and brain isn't functioning right...

In practice, I suspect it's the case that the sample size of most digital images exceeds the output devices they are sent to by some margin already, so in practical terms there's no great problem.

 

[2F/2F]

Digital is like having 3 separate images on 3 sheets of paper.

IMO, this is the key to understanding it. Well put.

 

[tim_walls]

IMO, this is the key to understanding it. Well put.

Although we have to ignore the fact that in all commonly available digital cameras* the three images are all of different parts of the picture - thanks to Bayer filters - and we have to do the digital equivalent of squinting at them to make them blur into each other to actually see the finished product.

Not strictly relevant to colour spaces per se, but bears mentioning .



* I believe the honourable exceptions are some Sigma cameras in the 35mm space, which use a more film-like arrangement where the three light sensitive layers are stacked on top of each other, rather than having a Bayer-type arrangement. Also exceptions of course are scanning backs that make multiple passes for each colour or systems with one-CCD-per-colour. (I've no idea if either of the latter actually exist - but I'm guessing even if they don't at 35mm-equivalent sizes they will at larger formats.)

 

[Q.G.]

Although we have to ignore the fact that in all commonly available digital cameras* the three images are all of different parts of the picture - thanks to Bayer filters - and we have to do the digital equivalent of squinting at them to make them blur into each other [...].

so that two of the images are partly constructed out of the data in the third image.

But ones constructed, the three can be treated individually.

 

[mtjade2007]

In digital the primary colors are discreetly captured by individual pixels...

This is not really true. Digital sensors actually capture only two colors. The 3rd is derived by interpolation.

 

[Photo Engineer]

Digital sensors have 3 color sensitivities, R/G/B but the ratio is not 1:1:1 exactly. The lateral offset is compensated for in software but still causes aliasing and that is compensated for by the use of fine grids.

PE

 

[mtjade2007]

Actually most digital camera CCD sensors have only one CCD cell for each pixel and there is only one filter on top of each pixel. But R, G and B filters are arranged in a way all over the pixels that data for other colors are derived by interpolation from neighboring pixels. It is not like each pixel has its own R, G and B filters. Digital camera CCD sensors need resolution desperately. They do not use 3 cells per pixel each with its own RGB filter. On higher end film scanners the linear CCD sensors have 3 CCD cells with R, G and B filters on top of them for each pixel. This is why film scanners can capture more accurate colors than digital cameras. I believe this is why scanning slides often yields better images than digital sensors.

Here is some information about digital camera CCD sensors: http://en.wikipedia.org/wiki/Bayer_filter

 

[nicefor88]

If you take only a few pictures in artificial light with an outdoor film set at 5500 Kelvin then you can use a blue filter to compensate the yellowish dominance this will produce. But if your intention is to expose 1 roll or more indoor why not go for a special film created just for that purpose?

 

[srs5694]

if your intention is to expose 1 roll or more indoor why not go for a special film created just for that purpose?

Because your choices are extremely limited. Offhand, I know of no tungsten-balanced C-41 films, and the only tungsten-balanced slide films I know of are pretty slow (ISO 64 or thereabouts), which means you'll need a tripod in most situations. You could go for a tungsten-balanced ECN-2 film (like Kodak Vision 500T, if I recall the name correctly), but that's got its own problems, like difficult processing and contrast that's wrong for optical printing onto RA-4 paper.

 

[benjiboy]

Because your choices are extremely limited. Offhand, I know of no tungsten-balanced C-41 films, and the only tungsten-balanced slide films I know of are pretty slow (ISO 64 or thereabouts), which means you'll need a tripod in most situations. You could go for a tungsten-balanced ECN-2 film (like Kodak Vision 500T, if I recall the name correctly), but that's got its own problems, like difficult processing and contrast that's wrong for optical printing onto RA-4 paper.

Fuji Pro 160 L (NPL) is a tungasten balanced C41 Negative film available in 120, and large format sizes, I've shot loads of it in the studio with hot lights fairly recently.