Colour Calibration in a Converted Camera

[Andrea B.]

Standardizing the colour in a Visible photograph made with a converted camera requires two steps:

 

1) Application of a camera colour profile so that the camera's colours are properly described for translation into editing and printing colour spaces.

 

2) White balance made on a neutral reflectance standard, such as a white card or a grey card, to adjust the Red-Blue axis so that white, black and all greys appear neutral for their given amount of Green.

 

When a raw file is opened in a converter/editor, a stock camera colour profile is either initially selected or automatically applied. This colour profile describes the camera's colour curves and translates from the camera raw data to the converter colour space in use, typically ProPhoto. We need this translation in order to reproduce the colours that the camera actually recorded. Once that is done, white balance adjustments can be applied.

 

We note that the stock camera colour profile is built upon the camera sensor, the Bayer filter and the internal UVIR-blocking filter. There is some variation in stock camera profiles across the various converters and editors, so many photographers choose to create their own camera colour profiles, often taking into account a particular lens or light conditions.

 

A tool such as the Color Checker Passport and its associated software is typically used for creation of non-stock camera colour profiles. The converter Photo Ninja has its own CC tool for creation of new camera colour profiles. Other editing apps, such as lightroom, also have CC tools. Color profiling can be done independently of editing software, if desired.

 

Standardizing colours in a UV photograph requires the same two steps:

 

1) Application of a converted-camera colour profile so that the converted camera's colours are properly described for translation into editing and printing colour spaces.

 

2) White balance made on a UV-neutral reflectance standard.

For some converted cameras, white balance can be done in-camera. But after application of a camera colour profile and translation to the editing colour space, the white balance may shift and need adjustment. With other converted cameras, a complete UV white balance is only attainable in the converter.

 

Converted cameras require construction of new colour profiles.

 

After removal of the camera's internal UVIR-blocking filter, the stock camera colour profile no longer accurately describes the camera's colours. The fact that non-Visible light might have been used to record those colours is irrelevant. (To be fair, the converted camera's colours do not typically drift too far from the original colours, but it is noticeable in most photos if not adjusted for.)

 

So a new converted-camera colour profile must be must be built.

 

**********

 

I have only recently begun to seriously work out the use of converted camera profiles after wondering for a year why, in spite of our best efforts to standardize our UV colours, we were getting so much variation. It may not appear so to most viewers because the UV white balance step combined with stock camera colour profiles does produce a similar blue/yellow/grey look across all these UV photos. But I've look at hundreds and hundreds of them and see a lot of variations in those blues and yellows, particularly the blues. Seeing the same flower in both a cyanish-blue and a greyish-blue tells me something is not quite standardized!

 

**********

 

UV white balance standards in decreasing order of accuracy and expense:

 

1) calibrated Labsphere Spectralon reflectance standards (or other similar calibrated standards)

2) uncalibrated reflectance standards (from the same makers as make calibrated standards, or look on Ebay?)

3) PTFE disc (look for virgin teflon, not always easy to find, some may be more accurate than others)

4) Color Checker Passport black patch (quick & dirty shortcurt, less accurate of course)

5) White balance on a Magenta area in the converter/editor (quick & dirty shortcut, less accurate of course)

 

**********

 

ILLUSTRATIONS

 

This D700 Visible photo was opened in Photo Ninja and white balance was made on the reflectance standards,

but no camera colour profile was applied.

Photo Ninja's Plain Color Enhancement was chosen for its average saturation.

The neutrals are neutral.

The blues and the yellows are off as compared to the actual CC Passport.

 

The same photo with Photo Ninja's stock D700 "Daylight/Flash" colour profile

automatically applied upon opening the file.

Plain Color Enhancement was used again.

The neutrals remain neutral.

The blues and the yellows are better, but the reds are now off.

 

The same photo with a custom camera colour profile created in Photo Ninja with the Profile Light Source tool.

Plain Color Enhancement was used again.

The neutrals remain neutral.

Colours are where they should be. The saturation looks a little strong, but that can be adjusted further in the Color Enhancement tab.

 

**********

 

This D700 UV photo was opened in Photo Ninja and white balance was made on the reflectance standards,

but no camera colour profile was applied.

Photo Ninja's Plain Color Enhancement was chosen for its average saturation.

The UV neutrals are UV neutral.

The false UV blues are a little cyan-ish.

 

The same photo with Photo Ninja's stock D700 "Daylight/Flash" colour profile

automatically applied upon opening the file.

Plain Color Enhancement was used again.

The neutrals remain neutral.

The false UV blues are now a little purple-ish.

 

 

The same photo with the same custom camera colour profile applied that was used in the Visible example above.

Plain Color Enhancement was used again.

The neutrals remain neutral.

The false UV blues are now looking blue.

May we please keep in mind here that the important thing is not what shade of false blue we obtain,

but that we obtain the same shade of false blue consistently across gear and software platforms.

[nfoto]

Are your findings independent of the [output] colour space ? Just working in a wide gamut such as ProPhoto and then output to sRGB (or even Adobe RGB] will precipitate a lot of minor colour alterations.

 

I heartily recommend making a colour profile for the modified camera for use in visible light in conjunction with a UV/IR blocking filter. Colours otherwise will be off and frequently found near impossible to process to acceptable hues. Andrea's suggestion is spot on and will help solve this particular issue.

 

However, once the UV bandpass filter is active in the UV-enabled camera and all visible (and, hopefully, also IR) light is blocked, we have in effect another set of sensor characteristics and I simply fail to understand why the new visible-light correction now applies. In fact, the entire shape and cross-over points of the transfer curves of the Bayer dyes now in effect were blocked off when the new visible-light profile was built. These characterstics (now entirely outside the visible range but seen by the camera) determine the actual output UV "colours". Were we given a UV version of the Color Checker then a sensible profiling (or rather, fine-tuning) would be feasible as well as worth its while. In the mean time, running PhotoNinja with "No profile" selected in the Colour correction module may be used. This equals assuming a perfectly "white" light source and will typically push the false UV "blues" to a less purple hue.

[Andrea B.]

.....once the UV bandpass filter is active in the UV-enabled camera and all visible (and, hopefully, also IR) light is blocked, we have in effect another set of sensor characteristics....

 

The camera's sensor, its Bayer filter and its raw RGB colour description never change.

 

Light enters the lens, passes through the UVIR-blocking filter, passes through the Bayer filter,

hits the sensor and is recorded in as raw RGB colour data (figuratively, in the RGB channels).

 

The unconverted-camera colour profile

• which describes the camera's raw RGB colours, and
  
• which translates between the camera's recorded raw RGB colour data and the editing colour space
  

is built upon the response of the sensor together with the Bayer filters, the internal UVIR-blocking filter over the sensor

and the microlenses over the sensor.

 

The Visible unconverted-camera colour profile is applied so that we can reproduce the colours the camera recorded.

 

Is this all correct so far??

 

OK, then....

 

When we remove the UVIR-block filter over the sensor,

we have NOT changed the camera's sensor, its Bayer filter, its microlenses or its raw RGB colour description.

They never change.

 

However, we have invalidated the unconverted-camera colour profile,

and the translation of the raw RGB colour data to the ProPhoto colour space now contains some mistranslations.

We are no longer seeing quite the colours the camera recorded.

Whether those recorded colours arise from Visible, UV, IR or any combo of the three is irrelevant.

 

How we create a new converted-camera colour profile may still be open.

I have taken a stab at it and it does seem to work, but all suggestions are welcomed.

[nfoto]

When we remove the UVIR-block filter over the sensor,

we have NOT changed the camera's sensor, its Bayer filter or its raw RGB colour space.

They never change.

 

Yes and no. This is indeed open to interpretation and probably the underlying cause of the slight disagreement.

 

Let me elaborate. It is true the physical system of sensor and Bayer filter does not change when the ICF in front of the light recording surface is removed. However, the anti-aliasing filter no longer exists. The register distance for the optical system is altered so lenses no longer focus correctly, but adding a neutral cover glass over the sensor will by and large take the mismatch away.

 

However, think what really creates the RAW colours or data in the RGB channels - this is the responsibility of the Bayer matrix dyes. In the visible range, the removal of the ICF has created a colour imbalance since we no longer have a filter shaping (or weighting) the sensor response so longer wavelengths in the visible are curtailed. The ICF appears bluegreen due to its spectral reshaping function. It also blocks UV and IR by design. Shooting a ColorChecker target in conjunction with a UV/IR blocking filter allows us to create a new visible light profile for the camera. Thus, we can recreate the same rendition of a subject in visible light to match the output from the non-modified stock camera. This is very useful as it allows the one and same camera serve dual purposes, either work in visible light or outside the visible spectral band.

 

We agree up to this point.

 

Now, to address the real issue. In the stock condition the camera (almost) never recorded light outside the designated visible passband. The imperfections of the spectral response of the Bayer dyes were inconsequential for the recorded RGB data because the "invisible" wavelengths (almost) never entered the system. In fact, the maker went to great efforts to prevent UV and IR to enter the sensor. Over time the digital cameras got highly efficient internal blocking filters to this end. So the fact that the channel started to respond to UV below 400 nm because their spectral responses had one or more secondary lobe(s) was not important in the stock camera, but with the modified camera, these irregularities are what makes the picture as it were. Again, we should stress that these secondary lobes of each channel's spectral response never were a designated part of the orginal system - due to the spectral shaping functionality of the ICF. The irregularities were lurking behind the ICF barrier but never manifested themselves. That is why I designate the modified sensor as a new system in its characteristics. It has effectively new spectral properties and correction for the changes in the visible band does not entail a correction to be valid in the spectral range outside the visible band which is my point.

 

When we add a UV/IR blocking filter, and reprofile the camera, we have a facsimile of the original system. The spectral weightings were changed by removal of the ICF, but the reprofiling puts this into place again so we can shoot with confidence the recorded colours look normal. What happens inside the camera with wavelengths outside the visible range is of no importance because the camera is blind to them. This is essentially the identical situation for the stock and the modified camera alike.

 

However, once the visible band + IR is removed by our UV bandpass filters, the situation again changes dramatically. The camera profile whether it is for the stock or modified camera no longer is valid. In fact, we have put the camera and the processing software in a position in which no calibration data exists. We can force the software to believe the illuminant (whether it is sunlight or flash) is "UV white" by selecting "No profile" for the light source in PhotoNinja. Alternatively, we can keep a Daylight/Flash source since this will add a little warmth to the colours. In fact seen only within the UV band, sunlight or flash is "warm" because the longer wavelengths dominate over the shorter ones. Then, we can use our UV-neutral targets to put the processing software "on track" again. As long as the [residual] spectral response curves of the Bayer dyes within the UV bandpass filter's range are different, UV "colours" will result. They are of course false as by definition no colours exist outside the visible band, but that is immaterial for our purpose. However, only when all three channels contribute significantly within the UV passband the processed "UV white" will appear white to the human eye. Otherwise, we may get a prominent colour cast from the dominant channel with the UV range covered. For a broad coverage in UV provided by say a Baader U (Venus) filter, it is unlikely a natural subject will be all UV grey (or white), whilst a narrow UV bandpass filter will balance to say pink, blue, green, cyan, or yellow.

 

For the ultimate fine-tuning of the colour rendition in UV with our modified camera, we would require a UV target equivalent to the ColorChecker or similar. Such a target does not yet exist. Thus we have to make do with UV white-balancing against spectrally neutral targets. Ideally, this UV balancing has to be carried out for each filter + lens + camera combination. Shoot a few frames that include suitable targets in the beginning of each session. I recommend a minimum of two targets such that a secondary correction can be employed later in the work flow if the conversion software doesn't get the colours properly balanced at first.

[Andrea B.]

I read every word carefully. Everything you write is true.

Indeed, everything you write, I have also written in one way or another - although our terminology may differ slightly.

************

 

Let us take as given the following points:

• Our goal is to produce a uniform UV-palette across the various UV-kits and converter/editors.
  A uniform UV-palette is necessary for documentary stability.
  

• White balance of our UV photos is a necessity for our goal.
  This was never an issue. We note that white balance, per se, is not a requirement for processing a UV photograph. We only care about it because we are aiming for that uniform UV-palette for documentary reasons.
  

• After converting our cameras by removal of internal filtering, we have a "new spectral system".
  This is well-described above. It consists of the filterless sensor, the Bayer filter and the camera software.
  

• The purpose of a camera colour profile is to describe the camera's raw RGB colour capability and to provide a translation between the camera's raw RGB colour data and the converter's RGB colour space, typically ProPhoto. Just agreeing on the definition here.
  

• Any light entering the new spectral system is recorded as though it were Visible light.
  I don't think this is an issue in our discussions. After all, there is no other way to record UV light using a Visible camera system other than in the camera's RGB colour channels. It does not matter what energy wavelengths enter the lens, only that once the energy does strike the sensor, the energy is dealt with as though it were Visible energy. The camera does not know it is receiving and recording UV energy.
  

• A photo file made by the new system must be handled as though it were a Visible photo file.
  I think it is this point where begins the hangup in our discussions. We have a Visible record of Invisible UV energy on our hands in the form of a Visible photo file. Everything we do to that photo file is done in the Visible RGB realm. We do not really have a UV photograph, we have a Visible simulation of a UV photograph.
  

Do we agree on these points?

If so, I will continue from my last point to once again try to make my case for a new colour profile.

[Andrea B.]

....guess no one is interested....oh well.

 

The next final point and the one I am most unsure about.

• The existing stock camera colour profiles are no longer valid for the new spectral system.
  Why? Because the stock camera colour profiles are built upon the old spectral system consisting of the filtered sensor + the Bayer filter + the camera software. My proof: Make a Visible photo of the Color Checker Passport with a converted camera and white balance it in the converter/editor of your choice. Are the colours correct? No. Why? Because a stock camera colour profile based on the old spectral system has been applied.

[JCDowdy]

....guess no one is interested....oh well.

 

Incorect assumption, continue if you please.

[Andrea B.]

I think I should finish Post #6.