Purple & digital sensors (long post)
Purple is one of the hardest colors to reproduce.
There are a lot of reasons for this.
First of all, I have yet to find a white balance system that handles well when pointed at a purple object - virtually all of them tend towards producing a solid blue, rather than purple. The only way around this is to white balance with a calibrated item like a proper gray card or an Expo disc.
Second of all, purple is a very hard color to capture in nature. Purple doesn't actually exist as a wavelength - violet does, but it's not the same color. Purple only exists as a combination of blue and red. Just from life experience, you should recognize that purples tend to change dramatically under different lighting - while a red is a red is a red regardless of the light, a purple may change to blue with a simple shift in light intensity or color.
Lastly, there are the technical details of the Bayer sensor. I'm a little muddy on this since it's been a year or two since I read a paper on it. Let me start here:
The light receptacles on a Bayer sensor (yours) only sense one color. Each light receptacle has a filter over it, and is sensitive to only one wavelength - red, green or blue. When light strikes the filter, the receptacle gets a charge, if that charge is very strong, the camera says, "that's a pure color" and reproduces it as such. If the charge is weak, the camera gathers the data around that receptacle and looks for other clues - if the green and the red sensors both got a little, it might be reproduced as orange. So that's how a Bayer sensor works (very simplified, of course), by looking at a bunch of light receptacles and figuring out how much of what wavelengths are there, then saying it's a certain color.
This runs into at least a couple problems with purple. Part of it is due to the Bayer pattern having twice as many green filters as red or blue - pure reds and blues render fine, and the green channel is reproduced faithfully, but there is simply more green data available than red or blue data, so combining red and blues doesn't always work as well, simply because there's much more green data surrounding the reds and blues, and purple has no part of the green spectrum. Having so much green data is actually a very good way to make sure colors are reproduced similarly to what we see, but it does lead to a limitation of purples.
It's compounded by the fact that violet is actually a shorter wavelength than blue and the camera cannot record violet properly. It records it as a blue. The blue filters are sensitive to the violet light, but it's still sensing it on the
blue
filter - there is no violet filter, and there is no combination of red blue and green that produces violet.
There's another facet about how the wavelengths of red and blue fit together (being opposing colors on the color wheel) but I'm not so familiar with the effects of that on the sensor.
Hopefully that's all somewhat accurate since I'm working on memory here, and someone can correct me about anything that's wrong.
Phew, that turned into something of a book. I hope at least one person muddled through it and found it useful.
Stephan
--
dev/null