What is the source of the differences observed between 24bit and 48bit color as generated by this system?

Process 1:

The scanner, as instructed by the software, scans a film image. No information about how this takes place is readily available, ie, the author does not possess the command set used by Vuescan to operate the scanner, nor is the manner in which that command set is used available. Vuescan is proprietary software.

Thus, it is not clear how the difference in bit depth is created. It could be a function of the scanner itself that is called upon by the software, or it could be that the software extracts what information the scanner can provide and casts it in the appropriate format. What is known is that the scanner itself is advertised as capable of delivering 16 bits of information on each channel. The software provides the option of utilizing either 8bit or 16bit information.

It seems reasonable to test the difference produced by Vuescan from a single scan, taken as a raw image and processed as both 24bit and 48bit color. For this test, the raw image will be taken in 48bit color; to expect the software to produce a meaningful 48bit color result from a 24bit scan seems pointless, and there does not appear to be any useful information gathered from the attempt. The point is open to criticism, of course.

It also seems reasonable to test the difference produced by the scanner by taking two different scans of the same image, one taken in 24bit color and the other in 48bit color, where the scans are identically configured and taken in immediate sequence. These sets of differences can then be compared.

Process 2:

These scans are saved to file in the TIFF format. Obviously, for these purposes, they should be saved in an uncompressed manner. At this point, we will have five files per image, four of which are intended for rendering and/or manipulation. The fifth file is a raw file, intended only to be reprocessed by the scanning software. It now remains to see what effect the rendering software has, when rendered as 24bit color and as 48bit color.

There are two rendering/manipulation software applications, and it is evident that they are part of the general question: they are quite similar, having a common origin as The Gimp 1.0.x, but one (The Gimp) does not utilize 16 bits per channel and the other (Cinepaint) does. (In fact, Cinepaint is reputed to be able to utilize up to 128 bits per channel.) These two applications are part of the reason for this investigation, and so their response to these files is relevant.

It is anticipated that a difference, to the extent a difference exists, will be evident in the different renderings of 48bit color files both by The Gimp and by Cinepaint. But at issue here is what difference is observed in the way each utilize both color depths. It seems reasonable, therefore, to submit each of four files to both applications to be rendered both as 24bit and as 48bit color.

This generates 16 renderings per image. Obviously, any two renderings can be compared. Of interest here, then, is what we might expect to observe. It seems reasonable to expect that Cinepaint will produce the greatest differences, because it is enabled for 16 bits per channel, where The Gimp is not. Perhaps the greatest difference will be between the files scanned as 24bit and 48bit color and rendered in Cinepaint appropriately, as these differences extend back to the original scans. Perhaps the least difference will be observed between the files that are scanned and those that are interpreted from raw as 24bit color by Vuescan when rendered by The Gimp. It remains for the consideration of the second question to make these determinations.

NB: Because nothing is known of the mechanics of these processes, there are no hypotheses that can be generated for these tests. They are procedural only, intended to sample the results of the different combinations of functions in the workflow process. Therefore, the question of the source of the differences, such that are found to exist, can only be assigned, not explained. Obviously, this is a source of significant question, and warrants further investigation and study.