On Tue, 17 Oct 2006 20:53:52 +0200, Dave Long wrote:
> (you all must have better phone cameras than I do.  I just tried
> capturing a PDF417 barcode, printed at about 0,6mm per pixel, and
> the resulting VGA photo is pretty poor, unless the 4-of-17 is sparse
> enough that it's unambiguous even when the the symbols are smeared
> into blacks, greys, and whites)

I think the NeoMedia "qode" guys suggest fitting an extra lens onto
the outside of your phone to change the focal length, which might be
one solution.  I haven't tried scanning PDF417s with cellphone cameras
myself, and UPCs are usually a bit blurry.

Barcodes are traditionally scanned with lasers, although the
LED-and-camera variety is becoming much more common.  The laser
approach gives you very good resolution in one dimension, without the
necessity to quantize up front, since that is a temporal dimension at
the scanner, not a spatial dimension.

If you designed a barcode to be scanned by a VGA camera at, say, 24
bits per pixel, it might look different.  Suppose you could ensure
that there was no rotational or perspective distortion --- each row or
column of pixels was exactly parallel to a row or column of pixels in
the cellphone --- but the pixels might not line up correctly and
therefore might be blurred.  It seems that in this case you should be
able to reconstruct the original barcode with Nyquist's sampling
theorem as a sum of sincs, at least up to the Nyquist frequency.  I
think this should allow you to get 24 bits per actual pixel, but the
barcode would probably look like multicolored spatial white noise
rather than a normal barcode.

I don't know how Fourier transforms change when you rotate things, but
I think it can't be worse than losing a factor of 2 in total
information, assuming you can estimate the rotation accurately.

Suppose you can only really get 2 bits per pixel, perhaps because you
have to deal with JPEG encoding, lighting variation, and printing
technology errors.  That's 640*480*2 = 614 400 bits, or 76.8
kilobytes.  (Commonly I get 2-4 bits per pixel for nice JPEGs.)  So
there's some headroom there, and it might even be plausible that you
could scan PDF417s.

What kind of programs fit in 75 kilobytes?  Most of the things I have
are either much bigger or much smaller, and lots of them are weighted
down with all sorts of JPEGs and stuff.

On Thu, 19 Oct 2006 16:49:14 +0200, Dave Long wrote:
> > - most of what they're using barcodes for could be done just as easily
> >   by typing four-digit codes on the phone keyboard.
> 
> Presumably a barcode can use all 10000 symbols, but manual entry would 
> require some redundancy to catch fat-fingering of fields.

A barcode could encode more than 10000 symbols, but the sample form
they showed only had ten or so fields.

> > - It might be feasible to put actual applications into a barcode rather
> >   than just field identifiers.  PDF-417 barcodes can hold several
> >   kilobytes of data --- I think up to 6 or so.
> 
> Not closer to 1K, especially with ECC enabled?
> 
> QR code claims to get up to almost 3K:
> http://www.denso-wave.com/qrcode/qrfeature-e.html

Thanks for the correction --- I hadn't looked at PDF-417 stuff for a
few years.


Reply via email to