16 or 8 bits per channel?

I have a Canon Digital Rebel and I normally capture my images as RAW. At
the moment, I have to convert them to TIFF to edit them in Photoshop. I
have the option of converting to 16 or 8 bit/channel TIFF. Is there any
benefit to using 16 bits/channel? I've always thought that digital cameras
only capture 24-bit color (8 bits/channel). If it would make a difference
with the print quality when printed to a digital printer at Wolf Camera or
similar?

Neal

 

It's a matter of where and when you choose to do your
most aggressive tonal corrections. To wit: if you do them
in your RAW conversion program, then save to 8 bits per
color, that works fine. But if you choose to defer those
corrections until later in the game, then have your RAW
converter output 16 bits per channel.

There's no printer in existence that can print using
16 bits per channel, so sooner or later you have to
convert the image down to 8 bits per channel.


rafe b.
http://www.terrapinphoto.com

 

What then might be the purpose of having a scanner that does 16 bit per
channel scans? If your photo editing requires you convert down to 8 bits,
discarding effectively 8 bits of something, before printing, why not save
time and file space and just scan/save/print everything at 8 bits?

Jim

 

Canon RAW is 12 bits. You will retain all 12 bits if you import into PS at 16
bits. That will means less loss of quality if you manipulate colour, contrast
etc in PS.

 

16 bits per channels is mostly useful if you plan to do
or need to do some really drastic tonal corrections,
or several such corrections in series.

Most scanner drivers and firmware operate with 16 bit
(or greater) math -- so it makes some sense to do the
most drastic tonal moves there, in the driver.

I've never been one to claim great advantages for
16 bit image manipulation. I do my main corrections
in the scanner driver (or RAW image converter) and
work in 8 bits in Photoshop.

There are of course lots of purists who prefer to work
with 16 bit images for as long as possible.


rafe b.
http://www.terrapinphoto.com

 

From: "JIM"

'Cause printing is the last step ... you can much of your work in 16 bit mode
(pretty much ALL of it with Photoshop CS) and then drop down to 8 to print.
Good explanation here ...

http://www.creativepro.com/story/feature/7627.html

 

'Preciate that link Bill. Did not like what I read, however. Appears I
haven't done enough homework concerning digital image manipulation. Bruce
wrote his article in Aug 2000, so am not sure how much of it still applies.
Also, do not know if the Epson 2200 can print 16 bit tiff files.........mine
prints something(?); but it could be that photoshop 6 just downsamples
automatically to 8 bits before sending the info to the printer? Also, don't
like not knowing what is being thrown away with those deleted 8 bits.

I am concerned now that I may have 'degraded' some of my images through
various manipulations in photoshop. Was not aware that I was creating
"holes" in my histograms with each change.

Another problem I did experience with one photo: printed out about 10 copies
of a photo on 13" x 19" paper. Printed each individually; i.e., could not
load all 10 at once. Now am wondering if the difference I saw in some of the
prints was not due to photoshop downsampling to 8 bits each time and
discarding different sets of bits for each print? Had bluer water in a
couple prints compared to the greener water of the majority. Also, the bluer
prints had a sharper appearance of the main subject (color more saturated).
Posterization or grain effect, as reported by Bruce, was more noticeable in
parts of the photo also.

An Epson tech rep, BTW, says they have no problems printing '24' bit files
with the 2200, never mind 16??

Did someone mention 'plug and play' printing ease?

Jim

 

In message <>,

Not exactly; you lose precision on those 12 bits when you move from a
linear space to a gamma-adjusted one. There are more values possible in
16-bit gamma-adjusted than 12-bit linear, but they are not the correct
precise values necessary, so they are rounded off.
--

 

There's no practical loss. If you were to convert from 12-bit linear to
16-bit linear, a change of 1 in pixel value in the 12-bit data would
become a change of 16 in the 16-bit data. A 16-bit gamma-corrected code
distributes its available codes differently, but the worst case (least
resolution) happens in the highlights. The one-step difference between
4094 and 4095 in the 12-bit code becomes a difference of 7 in 16-bit
gamma coding (65528 to 65535).

The largest possible rounding error that can be introduced by the
process is one-half code step in 16-bit gamma, which is only 0.07 step
in the 12-bit code. That's completely insignificant compared to the 0.5
step quantization error inherent in the 12-bit code, even with
noise-free computer-generated images. In photographs or scanned film,
the image noise will dwarf both of these errors.

So, for all practical purposes, there is no loss of information, and no
rise in noise level, introduced by converting from 12-bit linear data to
either 16-bit linear or 16-bit gamma encoded data.

Dave

 

[snip]

What he means is 24 *total* bit files, at 8 bits per RGB channel. No
printer that I'm aware of can print high bit (>8 bit/channel) info, because
no OS can send it more than that. PS just dithers to 8 bit if you
accidentally try to output a high bit file.

But, hey, I've got a burning question for you if you actually know an Epson
rep that knows his stuff: ask him/her what the optimum resolution is for
printing: iow, what dpi does the driver send to the printer.

IF he'she knows, and IF he/she'll tell you, you can help solve one of the
world's greatest mysteries!

 

No home pc printer can. High end professional printers like the $100,000+
Durst and Lightjet do print in true 36-bit color.

 

In message <401ca1d2$>,

What makes you think that printers that use error diffusion with small
dots have a single resolution? I'm pretty certain that every
inkjet/bubblejet I've had in recent years interpolates the size of the
dithering space for whatever PPI you send to it. With my S9000, I did an
experiment where I made a bitmap to be printed at 1200 PPI, and pasted
the same shape at offsets of one pixel, horizontal and vertical. Under
the microscope, it was quite clear that it did the same thing, 1/1200 of
an inch offset.
--

 

If you want to see some practical illustrations of the differences
between 8 and 16 bit images, I have an illustrated discussion on my
web site. Follow the tips link on the home page.

 

From the hosts of your favorite essay:
http://www.luminous-landscape.com/reviews/printers/lightjet_vs_inkjet.shtml

 

I didn't say "single", I was careful to use the word "optimum." Here's my
point:

Regardless of file DPI, you throw a file with a fixed size at an Epson
printer and the spool size (ie, print file) is identical. That means it is
rezing up to a certain dpi, and rezing down to a certain dpi. If I send an
8x10 @ 300 file dpi, I get a print file of 73.8MB; ditto for the same file
at 100, 150, 360, 720, 1440, and even 2880 file dpi. However, if you toggle
the *printer* dpi from , say, 1440 to 2880, then the spool file grows by
about 70%, which makes sense.

So, my question reamins, what does it rez up to, and at what point does it
throw file information away? And, it very clearly does both depending on
the situation.

 

A little while ago, somebody in this NG (apparently a pro) argued that
the "sweet point" for Epson printers was about 300dpi; and that
exceeding this figure significantly could harm to the quality of the
print. Somebody else disagreed, and said that Epson can handle any dpi
size you throw at it.

 

That's my point. There IS an answer, but who knows it? There is a fellow
who posts here that claims to work for Qimage and says they reverse
engineered the driver to determine that everything is spooled at 720 dpi.
Others say they absolutely, positively know the driver spools everything at
360 dpi. Which is right, if either?

All I know is the spool file remains the same regardless of image rez, so
the driver IS choosing a number and interpolates up or down as necessary.
I'd just like to know and move on.

 

http://www.luminous-landscape.com/reviews/printers/lightjet_vs_inkjet.shtml

They haven't written much I've ever quoted, if anything. Their scans look
different even on sceen, the Lightjet looks dramatically better and they
confirm it is true 36-bit, exactly as I said, so you were clearly wrong.
Did you have a point?

 

SNIP

SNIP

There was a discussion (that went semi-off topic) about that on the
comp.periphs.scanners group a month ago. The subject line was: "Help
understanding PPI and resolution", and the discussion became relevant to
this thread with the second post by Kennedy McEwen in that thread:
http://groups.google.com/groups?hl=...eKNyZKLM789$&rnum=10
You can Google your way down to the conclusions from there.

Bart

 

How can I be clearly wrong when I didn't make any assertions?

http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&oe=utf-8&q=%
22george+preddy%22+luminous+landscape&btnG=Google+Search

Just that a credible source with published credentials has disagreed with
you. In contrast with this credible source with published credentials,
you continously make declarations about various subjects with no backing
evidence and without publishing your credentials. Most of us don't have
access to Dursts or LightJets, so we can't confirm your comments about
them. I posted to a credible reference that, although confirming that
those devices are 36-bit, also mention that high-end inkjets do just as
good a job.

Hey, and since you brought it up, how do you get 36-bit files out of the
current crop of digital cameras without interpolation?