Book People Archive

RE: Digital camera OCR in FineReader 8

I just upgraded to Abbyy FineReader 8.0 and started testing that feature. I
used the very compact Canon S50 [5MP f2.8(W)-f4.9(T)] to photograph an older
book that I would typically scan. I used this camera because I know that it
will fit in my pocket and would not be too obvious if I were to try to go
scan a book in a library.

FineReader suggests that you photograph the book at maximum optical zoom (3x
for this camera - however I have a long zoom camera that have a 10 times
zoom, and it would clearly be overkill to use anything more than 3x zoom).
FineReader's suggestion makes sense because it reduces the amount of
curvature that the text incurs since being further from the page
approximates the lens being equidistant from all portions of the page (by
contrast, if you are very close to the page with the camera - say 12 inches
- then you will create a great deal of curvature in the text simply do the
fact that the edges the page are much further away from the lens than the
center of the page). The flipside of this, however, is that the further you
zoom optically, the less light you allow for the shot (e.g. f-stop of 4.9 in
this case), and that degrades image quality of the photograph which
diminishes the accuracy of the OCR, so some trade-off is necessary.

I tried three tests with my set up, but these were done in a hurry and I am
sure they could be done better. In all of these, I shot TWO pages/faces of
the book at a time and these measured 8in x 6in. I also checked FineReader
scan options for "Straighten Text Lines" and "Split Dual Pages."

1) I first set up a tripod, in a mode similar to a copy stand, and
photographed a few pages of the book with room lighting (no flash). The room
that I was using was probably not as well lit as a library, and the shots
came out fairly dark. The OCR accuracy, as a result was not too good (~15 to
20 errors/1400 characters/page). I should also note that I did not have the
camera at the full three times zoom in these cases. The camera reports that
the f-stop was that 4.5 - but this was a benefit over full zoom.

2) I then tried a few shots by hand to see if I could take photographs in a
less obtrusive manner than having a full tripod set up. In this case I put
the book on the floor and pointed the camera down. I sat in a chair and
braced my forearms on my thighs while holding the camera with both hands to
try to achieve maximum stability (basically you're a human tripod). Again, I
did not use a flash. These were by far the worst case. I had the camera at
full optical zoom (so the f-stop was 4.9), and the shutter speed was
something like 1/4 of the second, which is too long for a clear handheld
shot. The OCR error rates in these cases increased dramatically (~50 to 300
errors/1400 characters/page) and in some cases FineReader skipped over
portions of text because it was blurry or just did not look like text of
FineReader (even though I could read it).

3) I then tried the exact same thing as the #2 case, except I use the flash.
Using the flash caused the shutter speed to be 1/60 of a second which works
very nicely. In this case, I had the best OCR results and these were similar
to that of using a scanner (~2 to 7 errors/1400 characters/page). The
funniest thing was that the easiest way for me to keep the book pages open
in #2 and #3 was to use my big toes to hold the edges to the book taught! It
looks funny but works really well.

The bottom line is that the camera you are using and the amount of light in
the room will make the most difference. I do not think this is going to be
so dependent on the number of megapixels, though that will matter, as it
will depend primarily upon the amount of light at the camera lets through
the lens (e.g. the given f-stop for your particular zoom) as well as the
amount of noise that your camera creates in darker shots. In my case I may
have been a little better off using a 2x or 2.5x optical zoom to get a
higher f-stop. If there were a lot of light in the room and I had a camera
with a much lower f-stop at a similar zoom (say f3.0 or 3.5), then I think
option #1 (using a tripod with no flash) would have been the best solution
(if you don't need to be stealthy). However, if there are limitations in
terms of the amount of light (and you have no control over this such as
being in a library), then I think that going to a flash will improve things
dramatically. However, shooting off a flash consecutively is also going to
draw some attention. In short, a lot of this is just going to depend on
knowing your camera and being a good photographer. In short, experiment and
see what works best for you.

Hope this helps, Shawn

P.S.  

-----[From the] Original Message-----
From: Lars Aronsson [mailto:lars@[redacted] 
Sent: Tuesday, September 20, 2005 7:26 AM
To: The Bookpeople
Subject: [BP] Digital camera OCR in FineReader 8

Has anyone tried the new version 8 of ABBYY FineReader? Among the new
features is "Digital Camera OCR",
http://www.abbyy.com/finereader8/?param=44782#f2

This sounds very useful and promising to me, but how does it work in
practice?