[sane-devel] Re: Re: Re: epson 3490 - transparency unit problem

[Levente Novák]

Hi, 

On Wed, 2005-11-23 at 20:13 +0100, Oliver Schwartz wrote:
> Hi,
> 

> Lineart and grayscale should be fixable. The aspect ratio is correct 
> as far as I can see - it's rounded up to be dividable by 8, as each 
> pixel uses one bit and the smallest unit to request is one byte, i.e. 
> 8 pixels. My guess is that 1600 DPI works fine for both lineart and 
> grayscale, so the only remaining problem is 3200 DPI in lineart and 
> grayscale. Currently the deinterlacer is used for grayscale but not 
> for lineart, it seems that both settings have to be adjusted. Can you 
> send me example scans for 3200 DPI for a lineart and grayscale scan?

I will send them you in private mail.

> As for halftoning: I've no idea how to fix it. If the windows driver 
> doesn't offer it there's a good chance that the scanner doesn't 
> support it at all. I guess it should be disabled.

What does the snapscan backend for halftoning? Are there halftoning
algorithms implemented in software or does it only pass a flag to the
scanner? I'm asking this because there is a kind of halftoning, though
broken. If it's done by the scanner, it could indicate then it is
capable of some kind of halftoning.

> > PS: will be 2400 dpi enabled in the future? And why does the
> > windows driver allow 2400 dpi but not 1600?
> 
> You'd have to ask the epson guys. My guess is that the scanner scans 
> at 3200 DPI if 2400 DPI is requested and the decimating to 2400 DPI 
> is done in software. 

Yes, it was also my guess. There is also decimation at 1600 dpi, isn't
it (and at all resolutions below 3200)? The decimation patterns are
evident for most if not all small resolution modes (I checked it last
time), only this decimation is maybe done in hardware instead of
software. But of course it is easier to throw away every other
column/line instead of one of four.

> It could be done in the snapscan backend as 
> well, but I'm not sure if it's worth the effort. I guess the better 
> choice (in terms of image quality) is to scan at 3200 DPI and rescale 
> the image in gimp.

That's true, but in fact this applies to all resolutions below 3200dpi.
The best image quality you get will always be by sampling at 3200dpi
then downsampling to the requested resolution with bilinear, bicubic,
Lánczos, etc... algorithms.

I can see one advantage in scans directly performed at 2400 dpi: several
printers have a resolution of 2400x1200 or 4800x1200 dpi, it could come
handy when using the scanner as a copier (OK, it does not happen too
often, maybe if one scans slides/negatives and print them directly).

One further question: do you know what the real hardware resolution of
the Perfection 3490 is? In the technical data sheet Epson states it is
3200x6400 dpi. These 6400 dpi are maybe attained with the stepper motor
(as was the case with my former 600x1200 dpi scanner).


Levente