Challenging color matching head scratcher

[BenjyvC]

I'm working with custom lighting to produce photography used in photogrammetry, there being two models of LEDs used in separate exposures of nearly identical rasters. I'm limited to manufacturing tolerances in how closely matched are the spectral output of both arrays, there being a slight shift in overall chromaticity between images illuminated with each set of LEDs. This poses a problem when I go to layer these exposures in Photoshop to apply a difference blend, something I intend to use to tease out specular reflections in one of the image sets. This only works when the the color matches perfectly in both sets. I've had these LEDs tested and was provided a set of text files containing the needed spectral data, the question now turning to how to use this data to filter out of each data set what's not in common to the other data set, allowing me to work with a lowest common denominator. I'm not worried about dumbing down the color, as the isolated specualar data is converted to grayscale before it's put to use in a shader tree in a game engine.

Any idea how this might work in Photoshop. I'd write an action script to batch process the image database, just need to figure out how to make this work on two images based on the data in the .txt files. Here's a sample of how the data appears in the text file, seems to present a wavelength of light in one column with a response by the colorimeter next to that, then a repeat, perhaps a second sampling? I can find out more about the format of the data, but any ideas regarding scripting to use this data is mucho appreciated. Many thanks!

Random copy/paste from the blue end:

450.750000	0.011463	450.750000	0.308293
451.000000	0.011402	451.000000	0.310317
451.250000	0.011283	451.250000	0.310724
451.500000	0.011365	451.500000	0.316683
451.750000	0.011431	451.750000	0.322293
452.000000	0.011274	452.000000	0.321619
452.250000	0.011134	452.250000	0.321378
452.500000	0.011314	452.500000	0.330399
452.750000	0.011009	452.750000	0.325282
453.000000	0.010978	453.000000	0.328142
453.250000	0.010910	453.250000	0.329946
453.500000	0.010683	453.500000	0.326856
453.750000	0.010537	453.750000	0.326125

[D Fosse · Accepted solution]

The problem with LED is that it's not spectrally uniform. It has sharp spikes and deep troughs at given wavelengths.

What you're up against is known as metamerism failure, and this starts already in the camera sensor. This is why non-uniform light sources like LED or fluorescent require individual/custom camera profiles in the raw converter to (hopefully) reproduce consistently.

In other words this isn't a simple question of white balancing or shifting white point, as it would be with continuous light sources like daylight or incandescent.

I'm not saying it can't be done, but to start with it would require very well adapted camera profiles in the raw stage.

[Mylenium]

Could be some form of ASCII LUT format. When formatted suitably, you might be able to use it in PS using a Color Lookup adjustment layer. Once you figure out the specifics, you probably could also convert it to an ICC profile for easier use.

Mylenium

[BenjyvC]

Right, some type of ASCII LUT, will check with the firm that ran the test about just what format this file follows. The trick then is to use a PS adjustment layer that references a LUT in a format PS is happy with, so maybe I'll need some kind of format conversion utility. If not within this PS forum devoted to scripting, any ideas on what other scripting forums out there might deal with this kind of issue? Many thanks for the input, most useful.

[D Fosse]

BenjyvC  wrote

I'm not worried about dumbing down the color, as the isolated specualar data is converted to grayscale before it's put to use in a shader tree in a game engine.

I just have to ask the obvious question: Could you use grayscale/monochrome just for the Difference blend? I mean, if all you want is to isolate the specular highlights?

[BenjyvC]

D Fosse, thanks for the thought, but I left out one detail in the workflow (not to burden the reader). There's a bit of post-processing that has to happen on the isolated specular image while it's still in color before conversion to grayscale at the end.

Regarding your first comment about spiky data and what this portends at the sensor level, your thoughts are dead on and I'm interested to pursue that level of control in the future, but for present purposes, this exercise is less concerned with absolutely faithful color rendering than it is with realism in virtual lighting, which isolated specular data makes possible. As is, I do shoot RAW, am clear that the conversion invites all kinds of inaccuracies, probably mostly in the shadows and highlights, but given I'm using 92 CRI 5700K LEDs, the white balance isn't offensive. The inability to isolate speculars is. Thanks.