RE: Page images of books (using a digital camera)
- From: "Harry Plantinga" <hplantin@[redacted]>
- Subject: RE: Page images of books (using a digital camera)
- Date: Wed, 7 Mar 2001 11:46:01 -0500
> A copy stand would help with both these issues.
> In the past I've used an old photo enlarger, as it was taller
> than a real copy stand, and could be used with a camcorder that
> wouldn't focus as close as a 35mm camera. To do a proper job, a
> macro lens would be ideal to focus on a flat plane, but I don't know
> of any digital cameras that accomodate one, save the Nikon D1.
Focus isn't a problem with the Nikon; I can even focus close enough to
digitize microform prints directly, without magnifying (!). The camera
can focus on a 1/2 inch square, approximately, with resolution sufficient
to blow it up to full-page size. So, it doubles as a handy pocket
microscope.
Yes, a light table would help with illumination, but a software solution
is easier (er, once the software is written). If I wanted to do
illustrations I'd probably be more careful about lighting. Also, microfilm
magnifiers are notoriously bad about providing even illumination.
A copy stand doesn't completely solve the geometric distortion problem.
For that, pressing the spine onto a piece of glass (scanning) is best.
Some have asked about image size, capacity, etc. If I turn off image
sharpening in the camera, each image takes 900-1000k in highest-res
JPG format. So, I can get about 50 on a 48MB flash card. Dumping
the pictures to the laptop only takes a few seconds with a PCMCIA
adapter.
Yes, the algorithm (which we're calling "delight") could be described
as a high-pass filter, but more exactly, what it is doing is adjusting
each pixel's intensity to levels set according to the distribution
of intensities in the local region. Conceptually, the darkest 5% are
assumed black, the lightest 50% are assumed white, and the rest are
interpolated.
There is a graininess in the output that we think doesn't belong, and
we're trying to figure out the cause. Also, the current implementation
is very slow and we're worknig on a faster one. Finally, the current
implementation only opens and saves Sun Rasterfiles; it really ought
to be incorporated into a package like ImageMagick when it's done.
That said, if anyone wants to try out the C++ code, I'll make it
available.
-Harry Plantinga