Luminous Landscape Forum
Raw & Post Processing, Printing => Digital Image Processing => Topic started by: dmammana on October 23, 2007, 07:10:14 pm
-
When an 8-bit image is open in PhotoShop it's possible to go to Image > Mode > 16 bits. The file size then more than doubles, but what is actually happening here? Is it changing to a 16-bit file, making the processing to follow more effective?
If not, what is actually happening?
Thanks in advance. --Dennis
-
It can only add zeros into the extra 8 bits.
The advantage comes when you do some processing. The intermediate image stages will accumulate data in the lower 8 bits, so you don't get rounding-off errors in the upper 8 bits.
Best to shoot Raw images and convert to 16 bit TIFF intermediate files or transfer to 16 bit Photoshop image directly. You'll get 12 bits of data, and possibly as much as 14 bits with some of the latest high end cameras.
-
This functionality is really for compositing where you just have to match the bit depth of the two documents. If you want to paste 8-bit data into 16-bit etc. You're not getting the original benefits of high bit (more than 8-bits per color) from the get go.
-
So unless the image was captured or scanned in a higher bit format, I'd do just as well by leaving them in 8-bit files. Which is what made sense from the beginning... glad you confirmed that. (If that is, indeed, what you confirmed!)
-Dennis
-
So unless the image was captured or scanned in a higher bit format, I'd do just as well by leaving them in 8-bit files. Which is what made sense from the beginning... glad you confirmed that. (If that is, indeed, what you confirmed!)
[a href=\"index.php?act=findpost&pid=148251\"][{POST_SNAPBACK}][/a]
No. The point is, it's better to have a high bit-depth original, and process with high bit-depth, but lacking such an original, it is best to promote to the highest bit-depth in the editor. The worst choice is to start with a low bit-depth original, and process in low bit-depth.
-
So I DO gain something by converting old 8-bit images to 16-bit mode before doing my editing?
Forgive me for being slow on this one, but I'm just not sure I see how one gets something from nothing...
Dennis
-
So I DO gain something by converting old 8-bit images to 16-bit mode before doing my editing?
Forgive me for being slow on this one, but I'm just not sure I see how one gets something from nothing...
[a href=\"index.php?act=findpost&pid=148256\"][{POST_SNAPBACK}][/a]
It's not what one gets out of nothing; it's the armor an image wears to keep from getting further smashed into posterization. Any changes to saturation, contrast, levels, curves, etc reduce the number of colors in an image. Starting from only 256 levels for red, green, and blue can easily result in far less levels after just a few operations. Starting with 32768 levels might still leave you with a few thousand.
-
If you convert an 8 bit image to 16 bits and do a lot of editing while in 16 bits and then convert back to 8 bits you will get a smoother levels plot than if you did everything in 8 bits. This indicates you get some advantage in doing this, but not the same as starting with a 16 bit image.
-
"If you convert an 8 bit image to 16 bits and do a lot of editing while in 16 bits and then convert back to 8 bits you will get a smoother levels plot than if you did everything in 8 bits. This indicates you get some advantage in doing this, but not the same as starting with a 16 bit image."
Only if you've got dithering turned on for the 8 to 16 bit conversion, which, in effect, adds a small amount of noise, thereby smoothing your histogram. That dithering can cause problems in matching tones, making it all but impossible to produce an exact numerical figure.
The only time where it's any kind of real advantage to converting 8 to 16 bit is when I'm going to add computer generated imagery - gradients in particular - to an existing image. Other than than, it's pretty much wishful thinking that it's going to help you in any meaningful way.
-
Only if you've got dithering turned on for the 8 to 16 bit conversion, which, in effect, adds a small amount of noise, thereby smoothing your histogram. That dithering can cause problems in matching tones, making it all but impossible to produce an exact numerical figure.
The only time where it's any kind of real advantage to converting 8 to 16 bit is when I'm going to add computer generated imagery - gradients in particular - to an existing image. Other than than, it's pretty much wishful thinking that it's going to help you in any meaningful way.
[a href=\"index.php?act=findpost&pid=148538\"][{POST_SNAPBACK}][/a]
No, you don't need any dithering to benefit. In 8-bit processing, several values can become one, which would still be separate if they were 16-bit, and could still be separated to distinct 8-bit final values with further processing.
-
Only if you've got dithering turned on for the 8 to 16 bit conversion, which, in effect, adds a small amount of noise, thereby smoothing your histogram. That dithering can cause problems in matching tones, making it all but impossible to produce an exact numerical figure.
AFAIK, dithering only happens when converting from 16 to 8 bits, not the other way around. And a better approach when converting up is to do a light noise removal pass immediately following up-conversion. The NR algorithem sees the low-order bits being zeroed as a type of noise, and will interpolate intermediate values for them. It's still not as good as having a true high-bit source, but much better than just working an 8-bit file in 16-bit mode.
-
"In 8-bit processing, several values can become one, which would still be separate if they were 16-bit, and could still be separated to distinct 8-bit final values with further processing."
How does simply converting an 8 bit file to 16 magically spread those data points apart? You're still manipulating the same data, there's just more empty levels between your 8 bit data points.
-
How does simply converting an 8 bit file to 16 magically spread those data points apart? You're still manipulating the same data, there's just more empty levels between your 8 bit data points.
You've missed the point. Editing in 16-bit keeps values separated that would otherwise be mapped to a single output value if you stayed in 8-bit mode. If you do a curve adjustment on an 8-bit file in 8-bit mode, you may go from 256 level values in a channel to 150. Doing the same curve adjustment in 16-bit mode keeps you at 256 level values.
-
"In 8-bit processing, several values can become one, which would still be separate if they were 16-bit, and could still be separated to distinct 8-bit final values with further processing."
How does simply converting an 8 bit file to 16 magically spread those data points apart? You're still manipulating the same data, there's just more empty levels between your 8 bit data points.
[a href=\"index.php?act=findpost&pid=148662\"][{POST_SNAPBACK}][/a]
Try this: make a gradient across an image, from white to black. Then duplicate the image, and convert the duplicate to 16-bit. Open the levels tool for each, once to make the output 0 - 3, and then open it again to make the input 0 - 3. Compare.
A bit extreme, but it demonstrates what can happen.