The art of colour separations

A couple of years ago I browsed the stream of a fellow at flickr and saw his experiments into colour separations. I didn't think more of it until I browsed the online exhibition of Sergei Prokudin-Gorskij (1863-1944) at the U.S. Library of Congress called The Empire that was Russia. To make a long story short, Prokudin-Gorskij took three exposures of a scene through a red, a green and a blue filter. When these were projected afterwards it was possible to view a colour image. Nowadays (with digital image processing etc.) it's quite easy compared with then to scan each individual frame and then put together a composite image ('making of' section at LoC).

Generally I've seen very little of this on the net, at photo.net there was a post about a chap who lit a stilleben through RGB-coloured stage lights in order to get a similar result.

This I've attempted a couple times before with varying success. Here's my procedure:

Colour separation procedure

1.Take three exposures through a red, a green and a finally a green filter (in my case a 25A, X1 and a 80B as these are the filters I have [the letter and number combinations are the technical designations of the filters].l).

2. Scan them (to save time scan them in 8 bits/pixel) and set the image mode to RGB (if you scanned them in grayscale). Then tint them (in my case via photoshop's channel mixer, any equivalent would do) in red, green and blue (images #1, #2 and #3). For example, for the red image, using the channel mixer I left the red channel at 100% and set the green and blue channel at 0%.

3. Decide upon one image to use as the bottom one (or anchor) in order to line up the sequence of images (I've done as the LoC and chosen the red one). Then copy the green and blue image onto the red one. Set the opacity to 33% for the layer containg the green info and the same for the blue layer. Align both layers using the red one as a reference. Usually moving items (clouds, tree branches, waves etc. will stand out in either colour [RGB]).

4. When the images have been aligned, flatten the image. Usually one should see the embryo of the resulting image at this stage (in my case, red is a bit too dominant as evident by image no. #4). Then, using the levels or curves command or whatnot, set the correct colour balance, brightness & contrast. This is the tricky part. Of the images I've taken using this method only one has been somewhat suffessfull/accurate (see links below). As #5 and #6 show, one can muck about as much as one like to trying to get a pleasing result. Factors that influence the result is the spectral sensitivity of the film used, strength of the colour filters, the subjective treatment of the source files when scanning etc etc.

First attempt (red-green)

First RGB attempt (red-green-blue)

First successful RGB

A so-so sucessful sunset

More to come at a later date...