Hanging below may be best idea on this sub-thread.  Hanging below one could BACK LIGHT the paper, which would make even lighting much more easy.  A milky white piece of glass with several lights behind it, where if they were in front they would block the camera.  It would require compensation for the green bar fanfold, but maybe greenish lights would help that?

On 02/11/2018 02:10 PM, Timothe Litt wrote:

On 11-Feb-18 14:29, Davis Johnson wrote:
I think what you need is a wide carriage printer with the typical feed up through a slot in the bottom, and a camera.

The only working function needed from the printer is form feed. Photograph the page that is hanging below the printer, form feed and repeat.

Anybody here ought to be able to handle the programming to automate this process.

You would need to manually photograph the first page.

The camera would need good depth of field.

It's not that simple.  You need to deal with at least 2 common vertical pitches (6 & 8 LPI), and a number of page lengths (and widths).  These need to be setup per job; not all printers support all these.  Plus, misalignment (as Al noted, crossing the perforations at the bottom of a page is quite common).  The OP mentioned that his listings have a hard crease; this will cause (at least) feed and stacking problems.  Form feed causes a high-speed slew; this becomes less reliable as the distance moved increases.  You're proposing an entire page at a time - which means that the paper will jump off the tractors frequently.[1] Old paper is fragile.  Over hundreds of pages, dimensions may not be stable; it was not uncommon to have to re-adjust TOF after a while.  There's a fair bit of error detection and recovery to work out.

Lighting is an issue, as is compensating for keystoning and other misalignments.  Most cameras don't have a standard remote trigger interface - one of the pointers I provided loads modified firmware into cameras from one manufacturer to make this work.  If you look at digital camera reviews, you'll see that the lenses have varying degrees of artifacts, especially at the edges.  So you need to find and zoom to an area that's relatively "flat" & doesn't need a lot of correction.  While depth of field will help, it also will result in apparent font size changes as paper sways forward and back.  If you stop that, you simplify the OCR - and don't need as much depth of field.

There are many backgrounds that need to be subtracted for OCR to work.  (Printer paper was notorious for institutional logos, as well as bars and other aids to human readers.)  Then there are the other issues mentioned in my earlier note.

It seems simple, but it is a P.roject.  That's a capital P. With a lot of roject to work out.

It's worthwhile, but it's not simple.  It's a pretty interesting hardware (and software) project.  I don't mean to discourage anyone who wants to work on it - but you need to go in with eyes open, or you'll end up very, very frustrated.

Thunderscan tried to scan line by line & retrieve grayscale; the challenges were piecing together the adjacent lines with pixel resolution.   The focal distance was constant because the camera was on a carriage.  The idea here is to capture a page per frame.  So the registration problems are quite different.  One could try the thunderscan approach; it would trade one set of problems xxx "challenges and opportunities" for another.

[1] In my experience, with many brands and models of tractor feed printers over many years.  Paper handling is really difficult to get right.

Simh mailing list

Reply via email to