[sane-devel] Debug output from Lexmark X1270 scanner

[Stéphane VOLTZ]

Le dimanche 30 juillet 2006 03:17, vous avez écrit :
> Stef,
>
> I haven't had a chance to spend any significant amount of time with it.
> I'm hoping to get to it in August when I have some vacation time. There
> is quite a bit of work yet to do on it.
>
> I'll start by stubbing out all the calibration routines and just make
> sure it still scans as well as the old backend. Next I'll slowly add
> back the various calibrations and get them working for X1100 scanners.
> The amount of time a scan takes due to backtracking is unacceptable on
> some of the scans I tried - we're talking about several minutes to do a
> small scan at 150dpi! You thought this was due to the shading

	To disable shading, you only have tocomment the code below the '/* apply 
shading coefficients */' comment. Another thing I couldn't do was comparing 
register sets at final scan, since I don't have access to the needed model. 
Maybe there is a couple of values (apart from gain/offset settings) that 
makes some difference.

> calibration, but I have my doubts. I'll take it out to see if the
> performance improves. The calibration scans seem to take a lot of time
> as well, but that could be due to the new method for finding home

	Calibration is taking about as much time than under windows. The regular 
backend is much faster by not doing this, but relies on hardcoded values that 
don't work for all scanners.

>
> Where did you get the calibration algorithms from (gain, offset, and
> shading)? Do you know if the algorithms will work for scanners with a
> CIS head like the X1100s? Some of the hard coded ranges look "funny" for
> my scanner...
>
> Fred.
>

	These algorithms are inspired by scanner behaviour under windows, as seen in 
usb logs. The funny values are results of numerous tests and thourough try at 
registers contents. However, they are only tested on one model, and have to 
be validated on others.
	Software shading is a complete guess, since this image processing doesn't 
give any hints in USB logs. Comparaison with raw data and final scans under 
windows clearly shows that the windows scanning programm tend to over-expose 
and do some low pass filter to make shading artifacts completly go away.

	So the current status of the experimental backend is the backtracking you 
have on your scanner, and the tunning needed for calibration. In the process, 
the algorithms may have to be changed.

Regards,
	Stef