Blazing-fast PCIe storage for Mac Pro Tower

Color Temperature & Noise with Digital Cameras

Last updated 2010-01-09 - Send Feedback
Related: none

When shooting RAW with a digital camera, the white balance may be ignored, because it can be selected during subsequent computer-based processing of the RAW file.  But is there a downside to this practice?  This article explores whether failure to use daylight-balanced lighting results in increased digital noise.

While color rendition under non-daylight-balanced lighting is another consideration, this article does not investigate that issue.  For the purposes of this article, “daylight” is considered to be 5500° Kelvin.  See the Test Procedure for details on how the images were shot. 

With digital capture, one might expect the same kinds of color balance issues as with film.  But in fact, digital is amazingly forgiving of color-balance errors, at least when working with 16-bit RAW files.

Observe the crops with their histograms below and note the strong color casts on the latter two.  To facilitate comparison, the various crops are also presented at actual-pixels size on a separate page against a black background.  Please consult that page for the easiest visual comparison.

The strongly yellow crop is deficient in blue, and the strongly blue crop is deficient in yellow (red+green=yellow), as seen in the accompanying histograms.  (Yellow and blue are complementary colors in the additive Red/Green/Blue color model used by digital cameras).  Click on each item individually for a full-sized image.

4100° K = 244 mireds (measured)


+130 mireds = 2670° K (B+W KR12)

-130 mireds = 8800° K (B+W KB-15)

Below are the same crops as above, but gray-balanced in Nikon Capture using patch #8.  Read on for commentary.

4100° K—gray balanced


2670° K—gray balanced

8800° K—gray balanced

Gray-scale rendition is excellent.  A shift of 260 mireds (130 each way) is a huge shift in color temperature, yet a neutral gray scale has been reestablished for all three crops, with no obvious problems.  This is an impressive result, and one unlikely to be matched by film.

Whether all colors in a scene would be rendered equally well after such a major adjustment is open to question, but the ability to produce a neutral gray scale suggests that color rendition should be excellent.  While this article does not evaluate color rendition under such a wide variance in color temperature, experience suggests that color rendition is in fact excellent.

Please note that light sources with an uneven “spiky” spectral distribution, such as fluorescent lights and mercury vapor lamps might behave differently than the straightforward yellow/blue shifts in color temperature tested here.

Noise

In the above examples, we’ve seen that even with a huge shift in color temperature of 260 mireds, we can still establish a neutral gray scale.  Yet the histograms show that for the warm (yellow) example, the blue channel is underexposed, and for the cool (blue) example the red and green channels are underexposed.

The question is whether the adjustment has come at the cost of increased digital noise.   Increasing the luminance of underexposed channels will amplify any digital noise in that channel, so we might expect to see a significant difference in noise, especially in the darker areas.   Experienced digital users may already be able to guess from the histograms that noise might not be an issue—but let’s prove it by example.

To make it easier to find noise, we’ll examine the darker areas of the test strip.  With the Lab color space, the luminance values of patches 13-21 are {31, 25, 18, 15, 10, 6, 4, 2, 1} (the value 100 being pure white and the value 0 being pure black).  These values were obtained from the original file; the sRGB versions shown here might differ slightly.

When viewed under bright light, prints made on the Epson Stylus Photo R2400 using Premium Glossy paper show tonal separation between patches 19 and 20, a rather impressive printer performance.  Please note that this contradicts some claims seen online that luminance values under 10 can’t be distinguished from black—in fact steps 17-19, with luminance values {10, 6, 4}, show clear tonal separation even under moderate lighting.

There is no discernible difference in noise with these crops at their natural brightness level, either on screen or in print.

Actual pixels crops
4100° K
2670° K
8800° K

Below are the same crops, but brightened considerably.  The luminance values now range from 64 (brightest) down to about 18 (darkest), a large increase in brightness.

There do appear to be differences in noise, but they are so slight that comparing different-numbered patches leads to different conclusions; compare patches 16 and 17 using the black background page.   Any minor variation at such extremely low luminance values becomes subject to semi-random variations in noise, as well as very slight exposure variations.  In short, for all practical purposes, the noise levels are identical.  A test print on the R2400 confirms this.

Actual pixels crops, considerably brightened
4100° K
2670° K
8800° K
MacPerformanceGuide.com

High ISO considerations

One might ask whether higher ISO values show an effect.   But the question is specious: both the “cool” and “warm” exposures required an additional 4/3 stop exposure.   Use of a higher ISO is justified when a higher shutter speed or smaller aperture are needed.   Adding filtration as in this example would require ISO to be raised by an additional 4/3 stop in order to use the same aperture and shutter speed.  Raising the ISO is a self-defeating proposition when noise is a priority.

Conclusions

Assuming 16-bit raw processing and optimal exposure, use of non-daylight-balanced lighting makes no discernible difference in digital noise on the Nikon D2x, even in near-black areas.   This is true even for extremely yellow light (2670° K) and for extremely blue light (8800°K).     Theoretically, there might be benefits to filtration under more extreme circumstances, but this was not tested.

Nikon D2X users can safely skip warming or cooling filters in all but extreme situations without worrying about increases in digital noise.  This result is likely to apply to other models and brands  of digital cameras.  Thus digital offers yet another advantage: there is no need to buy a set of expensive warming filters.

Please see other diglloyd articles and reviews at diglloyd.com.

Test procedure

Tests were performed as follows:

  • ISO 100
  • continuous, full-spectrum side-lighting measured at 4100K using four Solux bulbs.
  • Nikon D2X with 85mm/f2.8D PC-Micro Nikkor at f8
  • B+W KR-12 and B+W KB-15 filters
  • Stouffer R2110 step-wedge
  • Nikon Capture 4.4.1, Color temp = "Daylight", no sharpening, Tone=Normal, Saturation=Normal, saved in 16-bit ProPhotoRGB

The Solux bulbs were allowed to warm up for approximately 45 minutes.

The 85mm PC-Micro Nikkor must be stopped down manually with a plunger.   Use of this lens guaranteed that the identical aperture was used for all exposures, since the diaphragm was stopped down once, and not changed while the exposures were made.

Exposures were bracketed, and a frame at each color temperature was chosen such that the white patch on the step wedge retained detail (no color values were pinned to the maximum).  For both the “warm” and “cool” frames, an additional 4/3 stop exposure over the unfiltered exposure was found to be optimal.

The luminance values for the warm and cool examples were slightly lower than the unfiltered example, due to unequal values of the RGB color channels. The luminance values were increased slightly using 16-bit Lab in Photoshop.

The examples on this page were converted into sRGB in 16-bit mode for display, then saved as 8-bit maximum-quality JPEGs.

Addendum—Micro-Reciprocal Degrees (mireds)

This section may safely be skipped, but is provided as background for the interested reader.

A  micro-reciprocal degree, or mired, is calculated by dividing 1 by the color temperature (measured in degrees Kelvin) and multiplying by 1,000,000:

mireds = 1,000,000 / (degrees kelvin)
degrees kelvin = 1,000,000 / mireds

Thus, a color temperature of of 5500K (which most color films and digital cameras consider to be “daylight”), has a value of 182 mireds.

Mireds are useful because they are a linear measurement, whereas degrees kelvin is non-linear.   A warming or cooling filter will warm (decrease the color temperature) or cool (increase the color temperature) of the light by some number of mireds.  To calculate the difference in mireds between two color temperatures, first convert each color temperature to mireds, then take the difference.  You cannot add or subtract color temperatures—you must use mireds.

I have measured all of my B+W warming and cooling filters using a Gossen Color Pro IIIf color meter, and marked each filter with the number of mireds.  This is important when shooting film, so that appropriate filtration may be quickly selected to bring the color temperature closer to 5500K.

A B+W 81A warming filter is typically 15-20 mireds, and a B+W 81B is about 30 mireds.  For example, using an 81B filter with 5500K light yields (182 + 30) = 212 mireds, or about 4700K.

Suppose you’re shooting in the shade high in the mountain, and you know that the color temperature is 20,000° K.  How many mireds would be needed for daylight balance?  Calculate as follows:

20,000K = 50 mireds
5500K   = 182 mireds

Correction needed = 132 mireds (warming)

A B+W KR12 (deep yellow) warms by 130 mireds, which is nearly a perfect match.

Rigorously lab tested and OWC certified.

diglloyd Inc. | FTC Disclosure | PRIVACY POLICY | Trademarks | Terms of Use
Contact | About Lloyd Chambers | Consulting | Photo Tours
RSS Feeds | Twitter
Copyright © 2008-2017 diglloyd Inc, all rights reserved.