Color Space Page 5
The bottom line: Editing steps are dangerous. You can improve things, but you also run the risk of turning large areas of gradually changing colors into areas with visible discontinuities (remember the "bad sky" illustration a few pages back). What to do?
Answer: Use a color space that has finer steps within the important colors, so that 1/128 of the total grayshade range is less perceptible. That's why there is such a thing as Adobe 1998 color space, and it's not a coincidence that Adobe 1998 color space was invented by a company making photo editors, and not by a company making cameras.
Q. Wait a minute. Can't I just do my editing in 16 bit color space? Doesn't that eliminate any issue regarding "just noticable difference" grayshades?
A. Yes! If you are using 16 bit files, Adobe 1998 has no advantage whatsoever over sRGB. But you will have to address these issues: (1) With 8 bit files you have a multitude of choices for image file format (e.g., jpg, bmp, png). With 16 bit files you are stuck primarily with Photoshop and Tif, both of which will take up a lot of room on your hard drive. (2) We won't accept 16 bit files. When you are done editing, you will have to then convert back to 8 bits when you are done.
Q. But won't the 8 bit files be of lower quality?
A. No, at least not noticably so. Do your editing in 16 bit color space, then change back to 8 bits when you are sure you are done editing. If you check the levels hystogram at that time, you will probably find that it looks just fine.
A. Yes. There are two important concepts: add noise and dither.
What we said earlier about the eye being able to perceive 200 grayshades needs to be said with a caveat: this is only true provided that there is a significant region at one grayshade, adjacent to another significant region, with a fairly regular boundary, with the other grayshade. If your file is prepared at around 300 pixels per inch at the size at which it will be printed, a fairly large amount of grayshade variation can be made on a pixel by pixel basis and not be noticable, provided that over a larger area (e.g., .02 x .02 inches) the average grayshade is what is desired.
For example, suppose that in one region you have a grayshade of 100, and in an adjacent region you have a grayshade of 101, and want to increase the contrast by 20%. Without noise or dithering, leaving the two regions at 100 and 101 is too little contrast change, but 100 and 102 is too much. Of course, if you had 16 bit files, you would be starting with 100.000 and 101.000, and be able to change them to 100.000 and 101.200, but in our example we're limited to 8 bit files. Let's say that before you increased the contrast, you added a small amount of noise (in Photoshop, there are varous algorithms for doing this), so that some of the pixels are at 98, some at 99, some at 100, some at 101, and some at 102. If your file is "res'd up" to around 300 pixels/inch at the at the size of the print you will be making, and you add a percent or two of noise, the result will not be detectable, even by the most critical observer! Then, when we increase the contrast, some pixels will have an increase in lightness by 2 grayshades, but most pixels will increase by only one grayshade. The average will be very close to just what we want.
The bottom line: By adding a small amount of noise to high-resolution files, you can, even with 8 bit files, get essentially all of the advantage of 16 bit files, but without the headaches caused by 16 bit files! Not only that, but now there's no need to worry about whether you are using sRGB or Adobe 1998 color space.
Continue to the next page to learn about dithering...