Hello all,
I am using DKL space as an isoluminant 2D color space, where the S-(L+M) and (L-M) axes correspond to chromaticity, and the (L+M) axis corresponds to luminance. My measurements are done with an ILT350 spectrophotometer, which measures illuminance rather than luminance. I do measurements in a dark room, at a fixed distance from the monitor. I have followed my DKL conversion explicitly after the matlab example in Human Color Vision, by Kaiser & Boynton (1996). Even with phosphor-specific gamma correction, I have had some trouble acheiving true isoilluminance.
Attached are two plots of predicted illuminance from the RGB coordinates of my (L+M)-constant DKL plane. Predictions are made with the spline model acquired from a modified version of the CalibrateMonitorPhotometer script which calibrates and fits each phosphor separately. I have also done predictions with the A_p*intensity_p^(gamma_p) models, which are less accurate. One plot is for a gamma-uncorrected space, and one is for a gamma-corrected space. The gamma-uncorrected space is clearly worse, but both are systematically non-isoilluminant. For the gamma-corrected space, the issue is worst at negative (L-M) values, where green is highest — this is intuitive since green contributes disporportionately to luminance.
I have developed a workaround illuminance-normalization solution which utilizes the A*phos_intensity^(gamma) model fit to illuminance measurements. I wanted to check here whether my approach seems rational and valid. The solution is as follows:
1) Acquire a gamma-uncorrected (L+M)-constant DKL plane and convert to RGB.
2) Predict luminance-output at each (x,y) value based on RGB coordinates, utilizing either the gamma or spline model.
3) Determine a scaling for each point, h(x,y) = mean(lum_out(x,y))/lum_out(x,y)
4) Determine phosphor-specific scaling h_p(x,y)=h^(1/gamma_p)
5) Multiply phosphor RGB(x,y) values by corresponding h_p(x,y)
6) Predict luminance-output again. If predicting based on more accurate spline model, iterate a few times to achieve better isoilluminance.
This works pretty well for me, and the resulting illuminance-normalized DKL plane looks roughly similar to the original, only clearly more evenly luminant (at least on my monitor). This approach has the advantage of not requiring explicit gamma-correction, which really washes out the space and also does not totally fix non-isoilluminance. I would like to stick with this approach, but have not heard of anybody doing something similar before.
Just wondering if anybody has any thoughts as to the validity of this approach. Will greatly appreciate any comments.
thanks and best wishes,
Nick Blauch
--
Nicholas M. Blauch, B.S.
Lab Manager
Computational Memory and Perception (cMAP) Laboratory
Department of Psychological and Brain Sciences
University of Massachusetts, Amherst