The old "8 x 16 bits" again...But worth reading!
2002-05-21 by André Vallejo
The following post is reproduced from a message by Dan Margulis in another web forum I join. I decided to reproduce it here 'cose even the subject has been discussed a lot here some time ago,it looks to me like a very enlightning about the subject ,written by someone who knows a lot about the subject (he's the author of "Professional Photoshop-A guide to color correction",one of the best books I've read). Sure it started a long thread in the forum,but that's the way it has to be...He gave me his ok for cross-posting. Andr\ufffd Shortly before leaving for Europe, I completed several days of work comparing the results of correcting 16-bit files versus performing the same edits on 8-bit files. I'll have a full discussion of what this showed in my book later this year and will include the files on the CD so that others can verify what went on. However, since there was one finding that came as a surprise and would cause me to change workflow in some cases, I thought I would let people know here first. As many of you know, there has been this ongoing debate as to whether there's merit in leaving files in the cumbersome 16-bit mode as long as possible on the theory that corrections will be more accurate. There are those who have strenuously argued that this is the only way to go, that there is a night-and-day difference in the results, and that people who do things any other way are rubes. None of these experts has ever shown anything other than a histogram to demonstrate that this is so, and my own tests have shown up to now no substantial difference. I therefore issued a challenge for those who could come up with any color photographic image where any reasonable course of events might conceivably show an advantage for working in 16-bit as opposed to just converting immediately to 8-bit, and said if I could verify it I would admit it and print the results. Naturally, none of the people who are so vehemently in favor of 16-bit correction had any such files to show, but some list members did, and sent them to me. Particularly, Todd Flashner allowed me the use of his scanner to produce several high-bit images that were disastrously flat and would presumably magnify any advantage that 16-bit might have. Ric Cohn sent over 2 gigs of images and an ungodly number of Epson proofs of images that he felt did demonstrate that working in 8-bit caused banding and other problems. The bottom line of all my tests was, with one important caveat that I'll get to in a moment, there is no 16-bit advantage. I blasted these files with a series of corrections far beyond anything real-world; I worked at gammas ranging from 1.0 to 2.5 and in all four of the standard RGBs, I worked with negs, positives, LAB, CMYK, RGB, Hue/Saturation, what have you. While the results weren't identical there were scarcely any cases where there would be detectable differences and in those one would be as likely to prefer the 8-bit version as the 16. So, I have no reservation in saying that there's no particular point in retaining files in 16-bit, although it doesn't hurt either. I'll show all these results later, but the surprise was in the files that Ric sent, which appeared to show just the sort of damage that 8-bit editing is supposed to cause, in an image with a dark rich blue gradient, a worst-case scenario in conjunction with the very dark original scan, which in itself was an attempt to give an advantage to 16-bit editing. Ric provided both original 8-bit and 16-bit versions of these files. Granted that the necessary corrections were very severe, they still showed that what he said was true: the 8-bit version banded rather badly and the 16-bit did not. I tried several different ways of trying to get around the disadvantage and could not do so without excessive effort. Ric's 8-bit original, however, was generated from the 16-bit scan not by Photoshop but rather within his own scanner software. Therefore, I tried further tests where I applied the same extreme corrections to the image, but this time not to Ric's 8-bit image but rather a direct Photoshop conversion of Ric's 16-bit image to 8-bit. Shockingly, this completely eliminated the problem. There was no reason to prefer the version corrected entirely in 16-bit. When Photoshop converts from 16-bit to 8-bit it applies very fine noise to try to control subsequent problems. Most scanners don't. I would have expected this to make a difference but not to the point that the scanner 8-bit file would completely suck and the Photoshop 8-bit file would be just as good as the 16-bit version. I don't know whether this is all a function of Photoshop's superior algorithm or whether the scanner is doing something bad. Furthermore, I don't care. One way or another, the 8-bit scanner file is bad and the 8-bit Photoshop file is good. I also don't know whether other scanners would have the same problem that Ric's appears to have, but suspect that they might. The whole thing suggests to me that if one's scanner is capable of generating a 16-bit file, one should probably take it. Thereafter, whether one converts it to 8-bit in Photoshop early or late, it doesn't seem to make a difference. Dan Margulis