How our Cameras see Color

Started Dec 30, 2009 | Discussions thread
ForumParentFirstPrevious
Flat view
Marianne Oelund Veteran Member • Posts: 7,780
How our Cameras see Color

It seems there is no end of mischief that I can get into, by simply adding a diffraction grating and optical slit to one of my cameras. Yes, it also requires a little programming, and lots of number crunching (my hard drive is full of massive spreadsheets now), but it's been a rewarding project.

An Unpleasant Surprise

Point a spectrometer at a light source with a smooth, continuous spectrum (such as an incandescent lamp), and you will see a nicely blended, continuous spectral display, which fades out gradually at the ends. Very pretty. I was expecting the same from the camera-become-spectrophotometer. However, the very first image from it was a bit of a shock:

Ugh - the red cuts out abruptly, instead of allowing orange to fade nicely into warm, then cool green. It's such a sharp break, that there is even a luminance notch around 565nm wavelength. Further, there is a wide band of pure green, in fact no hue discrimination at all from about 525nm to 565nm! To top it all off, the blue end looks like it's been whacked off by a Samurai.

As a sanity check, I reviewed the camera settings, and went to NL picture control with -3 contrast to bring up the weaker color components. Still not good, so I repeated the test with D3S, D3X and D300: Same result!

I decided it would be best to make a careful scan of the images, and plot the color channels as rendered in the JPEGs, for a start (Red channel shown in pale yellow - sorry about that):

That of course, only confirms what my eyes already told me. Still hoping for a better result, I sat down with my C compiler and produced a utility to read the spectral data directly from the RAW files. Viewing the data on a linear scale, without the saturation-enhancing function of a JPEG engine, we can clearly see how the filter bandpass responses are shaped (again, light source is a 50W halogen lamp):

There is definitely no denying that the red channel exhibits an extremely sharp cutoff at 570nm, and further, the blue channel cuts out abruptly at 420nm. The right half of the red curve looks as if Cookie Monster has taken a huge bite out of it; you can imagine a much more gradual slope connecting the primary peak with the smaller one at 530nm, but alas, it's missing.

After resigning myself to accept that our cameras are partially color-blind, I pressed on with the investigation, to produce a comparison of the D3-series cameras. Here are some composite plots, showing four cameras together, for each color channel (note the wavelength range is different for each plot). The data has been normalized (exposure-adjusted) so that the green curve areas are equal for all cameras. The differences you see in the red and blue curves truly indicate the sensitivity differences between the cameras.

The two D3 bodies, and the D3S, have similar but not identical red responses. The D3X is the weaker camera for red, especially in the way it dips suddenly past 660nm (left end); that makes it less useful for measuring spectra at the extreme red end:

The green channels of the two D3 cameras are amazingly well matched. All cameras track well on the left slope, but the D3S and D3X fall short at the peak, and instead have some additional response on the right slope. The D3X right slope is rather greatly extended, and one would expect this to interfere a little with blue saturation:

Again, the two D3's match well in the blue channel, and the D3S and D3X have lower peak sensitivity, with additional response on the slopes. It is interesting to compare the manner in which each cuts off at the 400-420nm end. The D3S and D3 #1 are extremely abrupt. The D3X and D3 #2 are slightly gentler, but the D3X really stands out for extending significantly further toward 400nm; it's the camera of choice for studying the blue end of spectra. From this, it can be concluded that the blue cutoff is a filter characteristic (perhaps a UV filter), and is not simply due to the lens that I was using:

Speaking of lenses, there is now lens color cast data, and some sample lens spectral plots, on my lens T-stop thread at
http://forums.dpreview.com/forums/read.asp?forum=1030&message=34091681

In all, color is an immense topic, and this is only an introduction to one aspect. There is much more to come, so keep an eye out for further posts to be added . . .

ForumParentFirstPrevious
Flat view
Post (hide subjects) Posted by
MOD TOF guy
MOD TOF guy
MOD TOF guy
MOD TOF guy
ForumParentFirstPrevious
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow