[argyllcms] Re: Printtarg dithering for TIFF output request

  • From: Nikolay Pokhilchenko <nikolay_po@xxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Thu, 19 Apr 2012 12:14:20 +0400

Thu, 19 Apr 2012 13:02:09 +1000 от Graeme Gill wrote:

> Nikolay Pokhilchenko wrote:
> > Well, it's a good question. I didn't want to explain to the end users why I 
> > ask them to print
> > 16-bit targets when they prints 8-bit images in general. I want building a 
> > good profiles with the
> > user's habitual image bit-depth. The end user can disable dithering by 
> > mistake. And if I send him
> > 16-bit target files, it would be impossible to characterize ink mixtures 
> > between quantization
> > levels. On the contrary, while printing at high printer resolutions with 
> > dithering, the
> > intermittent ink quantities can be characterized easily without 16-bit test 
> > charts and without
> > dithering at the user side.
> Hi,
>       I guess I'm a bit puzzled as to how this can work technically. 8 bit 
> dithering or
> screening is effectively linearly interpolating between the 8 bit values 
> using spatial
> averaging, and with a purely additive device this would lead to linear 
> interpolation
> between its 8 bit device output values, adding no effective extra precision 
> to the
> device measurement.

Yes, You're right. But the printing process isn't additive and (in almost all 
my cases) isn't linear. The number of drops of inkjet head between the 
quantization levels may be linear while spatial averaging. But in cases of high 
ink limits (quite often in my practice) or just steep device behavior (often 
too) the important breaks on device color behavior curves can be placed between 
the quantization levels of the printer. And it's necessary to measure 
intermittent levels of inks mixtures on the media. This can add an extra 
precision to process characterization.
I had a few cases when the dithering at a client side was disabled. When I've 
sending 8-bit quantized targets to the clients, the the quality of the prints 
with profile was unacceptable (without dithering at client side). After 
enabling the dithering, the quality was improved noticeably. But new profiling 
was necessary for the best results because steep printer behavior in a darks 
and in some blue region, which wasn't rendered by the target without dithering. 
From that time I've began generating the spacial dithered 8-bit targets without 
quantization and now I have no problems except asking the user to enable 
dithering if the problem arises at the profile exploitation stage.

> Apart from some very specific cases (such as instrument quantization and/or
> an 8 bit printing process such as dye sublimation etc.) the only way I could
> imagine this helping is if the linear light to perceptual characteristic is
> extremely steep at some points, and this might be viewed as flaw device model
> if it is interpolating in a space that is not a good model of the device 
> behavior.

I often deal with the printers which 8-bit device spaces are rather bad for 
perceptual characterization. Careful interpolation is needed including the 
spacial dithering between quantization levels of the device space. The 
interpolation wouldn't linear but it will allow to measure any point of device 

> I also guess that this is only with very specific test charts (calibration 
> charts ?)
> since typical profiling charts don't have enough patches to crowd them close 
> enough
> to be 1/256 of the gamut apart or less.

My typical test chart is 2880. If the device type is known, and I know the 
problematic, curved, steep regions, I generating special, additional path set 
(about 300 patches) for problematic region. Sometimes, if I have reliable 
profile for this device type, I'm generating perceptual spreaded additional 
patch set (with targen -I and -F parameters). So, the patch placing in 
problematic area can be very dense and justifies the dithering.

Other related posts: