CMOS Sensors - Blue Channel - Boosting

[Helen B]

The histogram is of a neutral target in 2850 K incandescent, as I mentioned in the description of it. It tells you at a glance that a CC30B would pretty much balance out the three channels in this and similar situations (should you wish to do so) - it's as simple as that, you don't need any other information. That is the point I'm trying to make. Where do you disagree?

 

[Caps]

BrianV

histogram data is a convolution of the three along with the colors in the image.


This discussion is getting a little beyond my beginners knowledge... but...

On the DXO labs site, they have reviews for the Nikon/Canon camera sensors. Does the information they give clear up the points you are trying to make?

 

[Helen B]

Helen, I don't fully understand your charts.

The light source you're using is quite light on blue spectral content, isn't it? So with a perfect sensor you'd expect the blue to be quite low. Is that what you're saying with the charts?

I'm having a little trouble distinguishing issues about the actual spectrum (light falling on the subject) from issues about sensor efficiency. Your charts seem to be about both, since they're describing what was measured by the sensor?

That is the whole point, and the direct simplicity of the method - they take both the sensor response and the spectral content of the light source into account. They also take the ADC's behaviour into account, of course.

It reminds me of the difference between Kodak and Ilford spectral sensitivity curves for their films. Kodak use a radiometric method (ie based on radiometric rather than light energy) and Ilford use a simple wedge spectrogram made in tungsten light. Ilford's method shows you roughly what to expect in practice, while Kodak's method needs further interpretation but is more versatile. As an aside to that: Unless you know the difference in method you would misinterpret the difference in spectral response, of course.

 

[amolitor]

I think the, uh, light is dawning. Maybe?

Can RawDigger predict what will happen with other light sources, or is it about telling you what's going on with THIS light source with great utility and precision?

 

[amolitor]

Helen, is the RawDigger chart, plus head-scratching and thinking, and then selecting the right filter essentially what "setting a custom white balance" does?

(albeit, by post processing the raw data on the way OUT of the sensor, rather than balancing the light going IN to the sensor, the latter process having some advantages, um, some of the time)

 

[BrianV]

The original post stated that silicon based sensors will always be noisy in the Blue Channel. That was true many years ago. Response in the blue channel was down 10x at 400nm with regard to a modern sensor, and the statement regarding noise in the blue channel was true. The tendency of the Blue channel to be noisier than the other bands has been eliminated. Then the discussion got off to using color correction filters as an advantage when lighting itself is biased towards one part of the spectrum. Using a color balance filter can alleviate this, but it is not worthwhile to undo the white balance algorithm of the camera. The histograms show the Blue response is about 40% the green and red, and the minimum value of blue was way above the noise floor. Use of a color correction filter is not required to defeat noise in the blue channel with this lighting source.

So if you use a calibrated lighting source, on a calibrated test target, without an external filter, you can basically back out the spectral response of the sensor. I prefer looking at data sheets to get the information.

 

[amolitor]

If you have adequate light, isn't a CC filter always going to be better? Doing white balance in post will always be about amplifying weaker channels which will amplify any system noise.

This is all rather academic, since clearly doing it in post is clearly good enough. I certainly don't intend to start correcting my light.

 

[BrianV]

If you want "correct" colors with a minimum of eyeballing it, would be nice to have a reference point. As shown, you can always back-out the data required.

I guess the use of a "Gain Flattening Filter" to even-out the light across all of the color channels would be the complement of the spectral response of the sensor and the light used to illuminate the scene. Even response across the channels to minimize noise. More important when balancing channels on a WDM network rather than photography.

Kodak used to sell a lot of film to the scientific/technical community, and it was used to produce radiometrically calibrated images. Truesense still targets the scientific/technical market.

 

[Helen B]

I think the, uh, light is dawning. Maybe?

Can RawDigger predict what will happen with other light sources, or is it about telling you what's going on with THIS light source with great utility and precision?

Yes, but it's easy enough to look at the histogram for a neutral target in two or three different light sources to get a more general idea of what filters would be required.

Helen, is the RawDigger chart, plus head-scratching and thinking, and then selecting the right filter essentially what "setting a custom white balance" does?

(albeit, by post processing the raw data on the way OUT of the sensor, rather than balancing the light going IN to the sensor, the latter process having some advantages, um, some of the time)

The use of physical filters to balance the raw values would be generally coarser than white balancing in post, not least because of limitations in the availability of filters, but fine tuning isn't necessary. It's a little like modern colour negative film. In the good old days we had to try to get a rough correction when using colour negative film, close enough for final correction during printing. As colour negative film became more and more capable of handling a wider subject brightness range the need for correction in camera became less and less important, if the exposure was adequate. The highest quality results are still obtained by getting an approximate balance in camera, but in many cases there may be very little difference between that and just fixing it all in post.

The original post stated that silicon based sensors will always be noisy in the Blue Channel. That was true many years ago. Response in the blue channel was down 10x at 400nm with regard to a modern sensor, and the statement regarding noise in the blue channel was true. The tendency of the Blue channel to be noisier than the other bands has been eliminated. Then the discussion got off to using color correction filters as an advantage when lighting itself is biased towards one part of the spectrum.

The inspiration for the original post appears to have come from a technical discussion on high-end modern cameras rather than cameras from many years ago. I think that considering the light source - sensor interaction is a more complete way of understanding this situation than sensor spectral response alone, but that is just my opinion.

Using a color balance filter can alleviate this, but it is not worthwhile to undo the white balance algorithm of the camera.

I think that if you are happy using the white balance algorithm of the camera then you are certainly not going to be bothered about this sort of refinement.

The histograms show the Blue response is about 40% the green and red, and the minimum value of blue was way above the noise floor. Use of a color correction filter is not required to defeat noise in the blue channel with this lighting source.

I'm not sure that you can make that deduction (in bold) from the evidence in the histogram. The histogram shows that the subject has a brightness range of merely a third of a stop for the majority of values, and there is only a stop between the extreme values - the minimum value of blue is only one stop down from the maximum value of blue, so one would hope that it is well above the noise floor. Normal subjects can have a much greater brightness range, and it is the combination of both brightness range and naturally lower blue signal that would determine whether or not the low blue values are above or below the noise/acceptable resolution floor. The acceptable blue noise criterion is going to depend on the use - there will be some post processing circumstances in which clean blue matters more than it does in general.