16 bits vs 8 bits' Thanks to Glen Mitchell
2004-01-04 by Julio Fernandez
Glen: What a great reply and exposition of the issues. That is a keeper, appreciate your taking the time to explain. Soon after I had posted I noticed your previous postings explaining why one should not confuse bit size with colour spaces. In this your later posting the explanations are deeper and I am delighted in your technical approach to the issues and awed by your technical competence. What a great resource for this group! . I should say also I was amussed about the apparent 'controversy' over Dan Margulis' statements. Guess that according to Margulis Abobe got it all wrong putting so much effort in 16 bit capabilities in their latest Photoshop CS. For my part I hope to soon buy Photoshop CS from the money saved not buying Margulis' book. Thanks again Julio Glen's Posting: <Message: 8 Date: Sun, 04 Jan 2004 06:52:49 -0000
Show quoted textHide quoted text
From: "Glenn Mitchell" <gmitchel850@...> Subject: Re: 16 bit vs 8 bit difference, Re Glen Mitchell\ufffds posting Thanks for the thoughtful reply, Julio. It is not the number of colors that is at issue. 256x256x256 is a large number of colors. More than 16 million colors. The problem is that the entire range of a particular color channel are distributed over 256 values. There are 256 discrete values for red, for example. Even in a smaller color space like sRGB, each one unit difference can cover quite a range of red. Now, as you work on the image, you move the information around. Adjacent pixels in a flower petal, for example, might start out one or two units different and then end up further apart. The result can quite easily be an image that goes from the appearance of continuous tones to one with visible banding. You only need to get non-continuous tones in one channel out of the three, where previously there were continuous tones to result in visible banding. With a 12 bit image, the same range of color for each channel is spread across 4096 values. That's 16 times finer available gradations. Same range for the information, but each unit increment covers less red, for example. This means you are less likely to see banding after making nonlinear changes. Smaller color spaces are indeed less likely to posterize than larger color spaces, but a 16-bit image in any color space is less likely to posterize than an 8-bit image in the same color space. (If you have posterization problems with an image in AdobeRGB or a larger color space, conversion to a smaller space can sometimes reduce/eliminate the effects.) The reason why smaller color spaces are less likely to posterize is that the range of colors is smaller. The 256 values in 8-bits and the 4096 values in 12-bits when spread across a smaller range means each unit change covers fewer tones. The gamut of the Epson 2200 printer broader than you seem to give it credit. It is not a small gamut at all. The gamut varies, depending on the paper and ink you use, but it is wider than sRGB and ColorMatchRGB, smaller than AdobeRGB (with the exception of a peak in the Yellow-Oranges). Don't get confused by the number of colors in 8-bits. That's really not the issue driving posterization. A change in an 8-bit image that results in just a one unit difference, say from 127 to 128, could be represented in a 16-bit image by a fraction of that same unit. The finer the gradations in color from one unit to the next, the less likely we are to see evidence of posterization. Cheers, Mitch Julio Fernandez