16 bits/channel (48-bit RGB, 16-bit Grayscale) VS 8 bits/c

Goldragon · Post by **Goldragon** » Mon Jul 21, 2003 1:43 pm

Hi I don't really understand the difference between 8 bits/channel (24-bit RGB, 8-bit Grayscale) and 16 bits/channel (48-bit RGB, 16-bit Grayscale)
?

Does it mean, 16 bits channel produces better quality result than 8 bit Channel???

Give me more details

Post by **NITeam** » Mon Jul 21, 2003 2:34 pm

In a 16-bit image, every pixel is represented with 16 bits (per channel). This allows to store two times more information than in an 8-bit image, where only 8 bits represent a pixel. The more information can be stored the higher image quality can be potentially achieved (of course this depends on the image source). In most cases, modern software works with 8-bit images but some professional tools allow to work with 16-bit images as well. Unless you already use a professional image editor that can handle 16-bit images, you probably will not be able to benefit from 16-bit output capabilities of Neat Image.

Hope this helps.

Vlad

jwarthman · Post by **jwarthman** » Sun Aug 10, 2003 7:25 pm

NITeam wrote:In a 16-bit image, every pixel is represented with 16 bits (per channel). This allows to store two times more information than in an 8-bit image

Vlad is right, in the sense that there are twice as many bits to carry the information in a 16-bit image.

But recognize that, at 8 bits per channel, a pixel can "only" have 256 "levels" of a single color. As a result, when making adjustments to the image, the colors often produce "bands" instead of smooth gradients.

But in fact, due to "binary math", each time we add a single bit we double the amount of information about the pixel. We can actually store twice the information of an 8-bit image in a NINE bit image! In 9 bits, we can store 512 levels! So, where an 8-bit image can store only 256 levels per color, a 16-bit image can store 65,536 levels per color - a 256X increase! "Black" is stored as "0", and a fully saturated color (e.g. red) is still stored as either 255 or 65,535 depending on bit depth. Can you see how banding and other artifacts can be dramatically reduced by having all those intermediate values when using 16-bit color?

When converting my RAW images to TIFF, I always use 16 bits per channel. I do as much processing as possible before converting to 8 bits per channel (for final sharpening and printing or saving for the web).

Hope This Helps!

-- Jim

*edited to correct my binary arithmetic!

Post by **NITeam** » Sun Aug 10, 2003 7:41 pm

Thank you for additional explanation, I hope this will help people who just start using 16-bit processing.

Let me correct one point:

Yes, a 16-bit image can have 256 times more levels but only two times more information than an 8-bit one. The latter is evident when you compare file sizes of two uncompressed TIFF files - a 16-bit and an 8-bit one that store images of the same pixel size. The 16-bit file is two times larger than the 8-bit one, which means it can store two times more useful information.

Vlad

Rob Keijzer · Post by **Rob Keijzer** » Fri Oct 31, 2003 2:04 pm

Another good visible effect using 16 bit is when you stretch contrast in an 8-bit image, then the histogram gets the shape of a comb. This is because you are pulling two adjecent levels apart.
Eventually this causes what is called here "banding".
with 16 bit you can stretch the contrast to twice its original range without this.
BTW IMHO this banding is called "posterization".

"There are 10 types of people: the ones that understand binary, and the ones that don't".

Rob

MAU · Post by **MAU** » Sun Jan 11, 2004 1:07 pm

Let me correct one point:
Yes, a 16-bit image can have 256 times more levels but only two times more information than an 8-bit one.
Vlad

Let me correct one point:

A 16-bit image can have 256 times more levels (and 256 times more information than an 8-bit one), but have twice more size.

WBR,
MAU

Post by **NITeam** » Sun Jan 11, 2004 1:23 pm

MAU,

"and 256 times more information" is incorrect. Amount of information is equal to or smaller than the size of (uncompressed) image file.

According to your logic:

- 100kb image file could store 1 image of some specific size
and
- 200kb image file would be able to store 256 images of the same pixel size,

which is obviously wrong.

Vlad

MAU · Post by **MAU** » Sun Jan 11, 2004 3:16 pm

to Vlad

Yes, you are right. Sorry.
65536 levels is twice more informative only, than 256...
It so.
WBR,
MAU

Thomas D. · Post by **Thomas D.** » Mon Jan 12, 2004 10:45 pm

Just another question on that:

My input files are 8-bit. Will the Pro-Version process these files with 16-bit and are these 16 bits then really be used, so that there is (at least theoretical) a quality improvement over 8-bit?

Thanks,
Thomas D.

Post by **NITeam** » Tue Jan 13, 2004 12:00 am

Improvement - probably not, better preserving of what is already there - yes. Overall, it is better to process in 16 bits than in 8 bits.

Vlad