• Ultraviolet Photography
  •  

How to make UV colours reproducible


4 replies to this topic

#1 nfoto

    Fierce Bear of the North

  • Owner-Administrator
  • 1,800 posts
  • Location: Oslo, Norway

Posted 20 January 2013 - 10:54

Rørslett, B. 2013. How to make UV colours reproducible. http://www.ultraviol...s-reproducible/


Once a decent UV capture is acquired, processing to get the best image quality from it becomes a major issue. In general, for digital UV use your camera's RAW options since significant processing may have to be conducted later. The first stumbling block is the colour balance and that forms the topic for this article.

By definition, colours are sensations that we perceive with our eyes in visible light. Outside the visible range, colours are not defined. They are outside their parameter definition domain and in a philosophic interpretation they therefore do not exist. In the visible range, we can with a limited accuracy make a prediction of the spectral properties of the object from its reflected (and recorded) colour. Thus, an object appearing as deep red should have high reflectivity around 650 nm. If the objects appear pure blue, it probably has high reflectivity around 450 nm. For a deep green object, we would estimate it to be highly reflective around 550 nm. However, our perception of colours is not reliable since we can consider a pure spectral band, or a mixture of different bands, to represent the same visual colour. Yellow is a good example of this. Mixing pure red and green in equal amounts will produce a yellow that cannot be differentiated from a narrow-band sample in the yellow range of the visible spectrum. Furthermore, the electromagnetic spectrum is linear whilst our colour perception is not. Magenta can be made by adding pure red and blue, but is perceived as being closer to blue than red thus combining the end points of the visible spectrum into a single colour. Thus colours of an object cannot be used inversely to estimate spectral property of the object, because the relationship is not 1:1.

However, does all this entail we should only render our UV images in pure black-and-white? To a human eye black-and-white is fundamentally the same as 'false' colours because our mind is a colour-processing engine. The tonality of different colours can merge into one another so they no longer are differentiated in b/w, or similar colours may be rendered differently as shades of grey. Basically we cannot deduce the original underlying colour or spectral range from a b/w image any more than we can from a false-colour or even true-colour rendition of the same scene. What we can appreciate, though, is the added resolution of fine detail that might be possible using colours, if the object(s) have different spectral properties.This leads to the inevitable idea that "UV colours" can be useful if they are predictable and reproducible from the UV captures.

The digital camera records colours through a device called a Bayer matrix. Essentially this is a grid of differently-coloured spots overlaid the digital sensor such that each pixel of the sensor has one and only one grid cell of the Bayer matrix associated with it. The Bayer grid cells are either red ( R ), green ( G ), or blue ( B ) from colour dyes added to them. Thus, each pixel is either R, G, or B. However, due to the nature of the Bayer matrix, each pixel has 50% G, 25% R, and 25% B pixels surrounding it. This colour mosaic is later converted into estimates of (R, G, B ) values for each pixel by a process called de-mosaicing. This works because scene colours tend to show spatial autocorrelation so for example, if a pixel is reddish then it is more likely the neighbours also have red in them. There are twice as many G pixels because the human perception of detail is highest for the green range of the spectrum.

What all the above boils down to is that outside the visible range, the camera may record colours on its own if the Bayer dyes respond differently to radiance in the "invisible" ranges of the electromagnetic spectrum. In practice, that means UV-A (400-320 nm) and partially UV-B (280-320 nm), plus near IR (700-1100 nm). Unfortunately, it is very common for digital cameras to respond to UV and IR in a similar manner, so one has to resort to bandpass filtration in order to eliminate the parts of the spectrum not being recorded. For UV images, IR contamination is a severe problem that has to be addressed and solved (reference: separate article).

The final step, after having acquired true UV image files, is processing them to a common standard. This of course assumes that results can be made independent of the gear that recorded the subjects in the first place. We shall not expect any resulting colour to be similar to what appears in the visible range of the spectrum, only that the colours do not depend on the specific camera used. Since UV images are susceptible to IR contamination, it is assumed an efficient IR-blocking filter is used so the maximum possible of the UV range from say 300 to 400 nm is available for the capture and everything above 400 nm is cut off by a factor of at least 103 (this in photographic terms is equivalent to a 10 stop reduction). Within the passband the filter transmission ideally should be perfectly flat at all wavelengths. In reality, these criteria demand cutting-edge filtration technology and both might not be perfectly manifested.

With this ideal UV setup in place, all that remains is to make a mapping so that the R,G,B channels respond equally for "UV-white" light. Again, like ordinary light sources for visible-light photography, the UV illuminants at our disposal are not perfect nor are they strictly "white". This means spectral energy is not evenly distributed across the entire range under consideration. What we can do in such a case is make a colour balance using spectrally neutral targets (reflecting evenly across the range of wavelengths we study). Preferably the target(s) should be Lambertian, ie. reflectivity is independent of the angle of incidence and accordingly, they also act as diffuse reflectors. Such targets are commercially available, but they are not inexpensive. White or grey pieces of PTFE (polytetrafluoroethylene, see http://en.wikipedia....afluoroethylene) plastics can be a cheap substitute but their Lambertian characteristic may be questionable.

Using such neutral targets, we can then achieve a profile or mapping such that an object with an even range of reflectivity over a wide spectral band records to the same level in the R, G, B channels of the camera, assuming the camera and filter combination itself captures a wide band of the spectrum, for example, the UV-A. Thus these targets will appear as white or in shades of grey depending on their reflectivity when such a profiling is conducted properly. Since we thus have balanced or profiled the camera's digital sensor to record in "UV-white", any object reflecting UV unevenly or having its reflectivity concentrated in a narrow band will be rendered in "UV-colours". These colours are of course "false", but as long as they can be reproduced reliably from one occasion to another, this is more a matter of getting better information from a UV record than proving or disproving the existence and reality of such colours..

Attached Image: CC_VIS_T1108117802.jpg
Figure 1. Visible-light rendition of a test scene. Flowers are Bidens ferulifolia (left) and Rudbeckia hirta (right). Both species are known to have strongly developed UV marks on their ray flowers. X-Rite ColorChecker® charts show the "true" colours of the scene. Two PTFE discs act as neutral targets. Nikon D3 (UV-modified), UV-Nikkor 105 mm f/4.5 lens. Baader UV/IR blocking filter, Broncolor Minicom studio flash with uncoated Xenon tube.

Attached Image: CC_UV_Clickwhite_T1108117822.jpg
FIgure 2. Scene as above and using the same camera, lens and Broncolor studio flash, but now filtration is by a Baader U 2" (Venus) filter. This ensures a true UV bandpass record. The camera's internal w/b is set to the lowest Kelvin possible about 3400 K using its Incandescent setting. Picture shown here after a standard click-white balance in a raw conversion program (Corel Aftershot Pro v.1.1.30, in this case). Most conversion software cannot make the necessary adjustments to the inter-channel balance by a simple w/b operation, so even though highlights may appear perfect, significant colour casts remain elsewhere. Different cameras handle this standardised "click-white" differently, but most cannot achieve anything close to the required neutral balance (but, see below).

Attached Image: CC_UV_Clickwhite_profiled_T1108117821.jpg
Figure 3. As above, but now using Picturecode PhotoNinja (v. 104c) and its Color Correction tool to set an appropriate colour balance. The desired neutral rendition of the targets are now apparent. An identical profile can be achieved by Aftershot Pro using curve tools on each channel, or by Capture NX2 using the local colour correction tool. So we can conclude that a preset profile or calibration applied in the processing pipeline of the RAW file can yield consistent results even though the profiling itself is carried out specifically for a given software application.

Attached Image: CC_VIS_in_camera_daylight_I1203103385.jpg
Figure 4. Another test setup with X-Rite colour charts and a set of Labsphere Spectralon Lambertian targets (spectrally neutral in the UV range). Daylight illuminance. Panasonic GH-2 (broad-band modified) with Coastal Optics 60 mm f/4 and the Baader UV/IR blocking filter.

Attached Image: CC_UV_In_camera_Daylight_I1203100752.jpg
Figure 5. Same test subjects, but now using Baader U2" (Venus) filter over the lens, and applying an in-camera white-balance done off the spectral targets. This, and other Panasonic models, can achieve a virtually perfect "UV-white" in-camera balance if adequate targets are used for the w/b. Again, we recognise the same set of "UV-colours" for the colour test charts to underline that these false colours may be false, but also reproducible, despite being obtained by entirely different equipment than in the first example of the Nikon D3. Moreover illuminants differed (flash vs. strong daylight), yet profiling removed this influence to give identical renditions.

All of the above relied on the assumption that the Bayer dyes did respond differently across the spectral window of investigation. This in turn means that with a narrow bandpass filter, this condition might not be met and thus instead of a colour palette we would end up with a monochromatic capture. Depending on the characteristics of the Bayer dyes in our narrow spectral window, the resulting colour might be any colour seen in the palette of a wide-band capture, or the resulting colour might even fall outside that palette. Again, using the colour as a look-up table to quantify spectral response in such a narrow band is questionable at best. Concomitantly, using narrow-band filters adds a lot of technical issues since exposures will be longer or demands for stronger UV light sources may be impractical or impossible to accomplish. Focusing can be more critical because of potential focus shifts of the optic in the narrow spectral band or if a live-view feature is used. Finally, when we descend into shorter wavelengths, the camera sensor usually is less responsive to UV as well. Thus one should have adequate reasons for going to the trouble of narrow-band captures unless there is a likely chance of getting something of special interest that couldn't be obtained otherwise. So far I have never seen a case in which such a scenario manifested itself, but that does not entail it cannot exist.

To illustrate the typical overall similarity between narrow- and broad-band UV captures, Figure 6 shows a compilation of UV images of Marsh Marigold (Caltha palustris; Ranunculaceae). Left, narrow-band 10 nm centred on 340 nm, right broad-band 300-390 nm with broad peak at 360 nm. Both have been colour profiled against spectral neutral targets. For the 340 nm capture, a monochromatic rendition is obtained, whilst the broad-band shows the usual palette. Comparing these two, one can readily see there is no difference in the UV patterns learned from the images. However, the price to pay for the narrow-band image is a great loss of depth of field since f/4 had to be used instead of f/11 for the broad-band capture. This results because there is a limit as to how strong a UV light source one can use, or in many cases, what the subject under study can tolerate before the photographic record itself kills it due to thermal output from the UV light source.

Attached Image: CALT_PAL_Narrow_and_broadband_I1205290246_UV.jpg
Figure 6. Both captures with Panasonic GH-2 camera and the Coastal Optics 60 mm f/4 APO lens, using Omega 340BP10 (left) and Baader U2" (right) filters, respectively. UV light source Broncolor Minicom studio flash (uncoated Xenon tube).

If we use the profiled broad-band camera for UV images we can get very subtle rendition of fine detail entirely lost by the narrow-band captures. Again, to show the nuances present in a stand of C. palustris see the next Figure 7.

Attached Image: CALT_PAL_I1206040373_UV.jpg
Figure 7. Panasonic GH-2, Coastal Optics 60 mm f/4 APO lens, daylight. Note shades of pale greens and blue hues in the foliage. These are seen identically rendered and spatially positioned on various captures conducted from different angles of incidence and direction.



[Published 25 Jan 2013 Last update 5 Feb 2013]


Bjørn Rørslett

#2 Andrea B.

    Desert Dancer

  • Owner-Administrator
  • 4,699 posts
  • Location: USA

Posted 23 January 2013 - 16:56

30 Jan 2017 Edit for clarification of phrasing in (1) about human eyes.

Remarks:

(1) Light of any kind enters a lens linearly and is linearly recorded by the digital sensor. Subsequently a gamma curve is applied by the internal camera software to produce a Visible photograph recognizable as "real" to our human eyes which, in conjunction with our brain, do not process light in a linear fashion. We are making the assumption that the camera's gamma curve should also be applied to our UV light captures. It is possible to convert a raw digital photo without the gamma curve application in either Visible or UV light, but I can see no particular benefit in doing this.

(2) Colour cast artifacts from the raw conversion process can adversely affect the UV colour standardization process described above.
Blowouts arise in UV photography just as they do in Visible photography - either from a specular highlight or from the camera's inability to handle a scene containing a wide dynamic range. Even if compensation is applied while shooting to control the bright areas, I have frequently found colour casts which appear after white balancing a UV photograph because of the extreme wrenching that is given to the colour wheel during this step.
Usually the colour cast manifests itself as a pure cyan area. But I've seen other colours also. I typically convert the cyan patch to white if the area represents a specular highlight. Sometimes if the blowout is due to lack of dynamic range, I will blend the blowout into the surrounding colour.
Added: Some of the problems with highlight control are converter dependent. Some converters are better than others at handling highlight control.

(3) Colour noise can also be difficult to deal with in UV photographs. Granted, all UV colour is false colour. But it is an interesting problem trying to recognize when some false UV colours are actually noise artifacts.
The backgrounds in most of our floral photographs are fairly UV neutral, so blotchy colours in a background are easily recognized as noise artifacts. But in the center portions of a flower where the UV flash can sometimes induce a bit of UV-induced Ultraviolet fluorescence (as opposed to the typical UV-induced Visible fluorescence with which we are all familiar) and where there are lots of blotchy areas of pollen or perhaps nectar, it can be interesting to decide whether one is seeing colour noise or UV data.

(4) Labsphere Spectralon Diffuse Reflectance Standards
http://www.labsphere...ds/default.aspx
I've been quite pleased with these calibrated, diffuse, Lambertian Labsphere standards. I would recommend that anyone pursuing scientific, standardized UV, Visible or IR photography make use of at least black and white reflective standards which are as uniformly reflective as possible across the UV-Vis-IR spectrum. The 99% white standard can be used to set the exposure white point as well as the white balance. I usually set the white point to be a little lower than than the 99% white standard.
With these Labsphere standards we have been able to determine that the PTFE discs are a reasonable alternative for determining white balance. Although the PTFE disc can sometimes go quite specular on you in really bright UV light.
Also we have found that the white balance cards used in Visible photography are not suitable for use in the UV spectrum. They can give colour casts which undermine the standardization effort.
Any of the grey 75%, 50% or 20% or Labsphere standards might also be useful for determining overall colour cast in a photograph when white balancing, but the camera's application of a gamma curve alters the presentation of their tones in a photograph so I do not use them for tonal adjustments.
The 2% black standard is useful for setting a black point. I usually pull the black point up a little above the 2% black standard.
Andrea G. Blum
Often found hanging out with flowers & bees.

#3 nfoto

    Fierce Bear of the North

  • Owner-Administrator
  • 1,800 posts
  • Location: Oslo, Norway

Posted 24 January 2013 - 22:16

The eye perceives light in a similar fashion to sound, so the response is inherently non-linear. The Gamma curve converts from linear (camera) domain to non-linear (human vision) domain, so obviously applies to UV captures as well.

Without the gamma correction, our impression of the world around us would be very dark and sinister indeed.
Bjørn Rørslett

#4 Andrea B.

    Desert Dancer

  • Owner-Administrator
  • 4,699 posts
  • Location: USA

Posted 24 January 2013 - 22:42

Application of a visible gamma curve to an ultraviolet photograph gives us yet another reason why we cannot attempt any reliable inverse mapping from a flower's UV photograph to its spectral properties, yes? Our UV photographs do not lie, but neither do they tell the complete truth.
Andrea G. Blum
Often found hanging out with flowers & bees.

#5 nfoto

    Fierce Bear of the North

  • Owner-Administrator
  • 1,800 posts
  • Location: Oslo, Norway

Posted 24 January 2013 - 23:02

They are in fact like any other photograph in that respect.
Bjørn Rørslett