Colour temperature actually shows where is the maximum in light emission. The lower colour temperature is, the more intensive is the red component. The higher is the temperature, the more intensive is blue component. We have sensor sensitive to all the visible spectrum (broader, actually); we are not mounting any filters on the sensor at this stage. In spite of the lack of filters, sensor will produce different response to the same amount of light - if the light has different wavelengths. Same, for different colour temperatures - same amount of light being a different mix of intensities of wavelengths will produce different responses. Now, let's mount the filter array. Let's say that filters are designed to match sensor response in a way that if gray target is lit with the light of 5500K, responses of sensors under red, blue, and green filters will be the same. Hence, we have reached "neutrality". If we raise the colour temperature of the light, bluish light components become more intensive. If total amount of light is the same, that also means that red is less intensive. Our gray target turns bluish, that is sensors under blue filters will generate higher signal then sensors covered with red filters. How can we bring our bluish gray back to neutral gray? We need to amplify red signal, and to quench blue signal. Vice versa for warmer light, we need to amplify blue and to dump red. Several thing from here. First, when this procedure is performed in digital domain, errors in white balance can cause overflow due to limited number of bits in the result and false highlight coloration. Moreover, correct camera white balance coefficients can play a dirty trick with such programs as Nikon Capture, where the two-stage approach to white balance is used. Capture first applies camera white balance data, and then the corrections user suggests. If camera white balance data resulted in loss of highlights (overflow), correction of white balance in Capture results in false colours in highlights, too. Because of this, one can say that invalid white balance results in exposure problems. Second, we do not need perfect white balance every time, as perfect white balance gives no hints of what the actual light in the scene was. Winter evening scene should be bluish, and asphalt at noon should be a little yellowish.
Iliah : That's a pretty solid explanation of the sensor and derived data system. Given that you don't want to start discussing sensitivity vs. wavelength for the sensor either mathematically or with a graph as well or similarly differential absorption vs. wavelength for the filters, it's a good way of addressing this. Here's a question, then. Does RML perform a "one-step" process with the data from the NEF/RAW file instead of overlaying a WB correction on the initial in-camera WB ? I'll assume that this would be why RML recovers more highlight detail. John P.
Dear John, but of course I can suggest graphs etc. We do that stuff routinly for RML. I'm just afraid that will go too technical. RML actually does two-step process, because we use floating point and we do not use multiplication as other converters. So, overflow is not an issue for us. Also, in new beta coming these days we will include some interesting suggestions.
Iliah : {chuckle} Well, that was what I was trying to say. I guess I just didn;t write my comment very clearly. Interesting. That explains a lot of things. Well, I'll look forward to the next beta. Maybe this time, I get into the swing of trying to use RM, as I missed the first time around. John P.
Dear Paul, No, there is no way to do that in a simple way For example, flash and daylight have very closely matching colour temperatures; but as spectrum is different, white balance is different too.
