bkrpr: book ripper/non-destructive scanner

[latchkeyed]

Originally Posted by chorpler

I found a local plastics company that will make the 12" version for $22, the 14" version for $27.50, and the 9" version for $14. But they're thermally bending the plexiglas to make the 90-degree bend. According to them, this will result in a "point" with an outside diameter of 1.5 times the thickness of the material, which is 1/8" plexiglas that has an actual thickness of 0.110", so it will be 0.165". Is this pointy enough to fit into a book and capture all the text, do you think? Is your plexiglass plate thermally bent, or did they join two pieces somehow?

Interesting. Most of mine are bent and I've never seen a point on the corner. Might have something to do with the bending process but mine all have round edges. I did get one built joining two pieces to see if that would make a big difference. If you go that route you'll want the pieces to be beveled together before joining and you will probably need some side pieces to support the joint.

As for the resolution, take a look at the OCR results here: http://bkrpr.org/doku.php?id=results

All those pictures were taken at 6MP. I've found 5MP is generally the floor for high quality OCR, but MP is certainly not everything.

[latchkeyed]

Originally Posted by latchkeyed

Interesting. Most of mine are bent and I've never seen a point on the corner. Might have something to do with the bending process but mine all have round edges.

I think I see what you mean now, a point of particular thickness, not a pointy joint. So yes, all of mine have been thermally bent and work just fine.

[chorpler]

Originally Posted by latchkeyed

Interesting. Most of mine are bent and I've never seen a point on the corner. Might have something to do with the bending process but mine all have round edges. I did get one built joining two pieces to see if that would make a big difference. If you go that route you'll want the pieces to be beveled together before joining and you will probably need some side pieces to support the joint.

Sorry, I was jusing using "point" as shorthand to mean, you know, the apex of the tent-like (or V-like) shape that the plexiglas forms. So yeah, it sounds like your "point" is bent and isn't a sharp point, just like mine will be.

[quote]As for the resolution, take a look at the OCR results here: http://bkrpr.org/doku.php?id=results

I noticed that -- is the "large" image the actual output from the camera? The letters are significantly smaller than a 300dpi scan, but FineReader 9 actually recognized your "large" pages from Free Culture with 100% accuracy -- it had about 10 questionable characters, but they were all correct. So maybe my scans don't need to be nearly as big as I thought. (Although FineReader did pop up a little message saying "Increase scan resolution for better accuracy with small characters." But its concerns were unwarranted since it still recognized everything on your page just fine.)

All those pictures were taken at 6MP. I've found 5MP is generally the floor for high quality OCR, but MP is certainly not everything.

Yes, I noticed in the thread you linked to above, about cameras, that an SLR camera with a 2/3" sensor would be much better than a regular $100 digital camera, regardless of megapixel rating. But since none of us want to pay out $1000 for a couple of used SLR digitals, we'll have to see what kind of cheap consumer digital cameras work best.

Oh, and another question: I notice in your video that the book is sitting on a V-shaped cradle. Is that one of the "do it yourself repro V cradles" that ereszet talked about in the thread you linked to? How much did that cost you to build?

Until I saw that video, I was mentally picturing the bookripper device as working the opposite way, with the book lying on top of the plexiglass and the cameras angled up at it. I wonder if such a system would work... or if you really need the weight of the device to hold the book's pages still.

[UncleDuke]

where have this been all my life??

[latchkeyed]

Originally Posted by chorpler

I noticed that -- is the "large" image the actual output from the camera? The letters are significantly smaller than a 300dpi scan, but FineReader 9 actually recognized your "large" pages from Free Culture with 100% accuracy -- it had about 10 questionable characters, but they were all correct. So maybe my scans don't need to be nearly as big as I thought. (Although FineReader did pop up a little message saying "Increase scan resolution for better accuracy with small characters." But its concerns were unwarranted since it still recognized everything on your page just fine.)

No, the image size is compressed by flickr. With a free flickr account you can't offer people the ability to download images in their original size. If you want the full images, grab them from here: http://bkrpr.org/doku.php?id=download

Yes, I noticed in the thread you linked to above, about cameras, that an SLR camera with a 2/3" sensor would be much better than a regular $100 digital camera, regardless of megapixel rating. But since none of us want to pay out $1000 for a couple of used SLR digitals, we'll have to see what kind of cheap consumer digital cameras work best.

Oh, and another question: I notice in your video that the book is sitting on a V-shaped cradle. Is that one of the "do it yourself repro V cradles" that ereszet talked about in the thread you linked to? How much did that cost you to build?

Until I saw that video, I was mentally picturing the bookripper device as working the opposite way, with the book lying on top of the plexiglass and the cameras angled up at it. I wonder if such a system would work... or if you really need the weight of the device to hold the book's pages still.

Yeah, we all live with what we have I actually built my v-cradle before I ran into the information here, so it is a little rougher than ereszet's. I just made it from some scrap wood and a lot of angle braces from the hardware store, so it was just a few dollars to build. I've been planning some different designs but it works pretty well as is.

You are not alone in thinking about a device where the book is on top, I talked to a couple of other people while coming up with this design and they also thought about flipping it upside down. Turns out there are a number of issues ranging from the book sliding off to being unable to reach the cameras, but it might be a good idea for a pre-manufactured device, like a scanner.

[kenbaldwin]

Very interesting! I've never tried scanning a book, but this looks like a fun project.

In the video, you're manually triggering each camera. Can't this be done remotely somehow? Also, are you recording the images to the camera card for later transfer, or directly to your computer via USB?

Hmm...after a little poking around, I found gphoto, which seems to do both these things, but unfortunately the Canon Powershot A590 you recommend doesn't support remote capture operation.

Ken

[kenbaldwin]

I downloaded the software and the test images. I was able to rotate and crop the images through the GUI, but I can't get it to work from the command line. I'm trying this:

./bkrpr.py -cr --odd 90 --even 270

but it's not recognizing the options. (Help is displayed.) What am I doing wrong? What's the proper command to process the test images? Thanks

[kenbaldwin]

Doh. Well, adding *.JPG to the command gets me past the help, but I get this error:

Traceback (most recent call last):
File "./../bkrpr-0.1.82/bkrpr.py", line 238, in <module>
main()
File "./../bkrpr-0.1.82/bkrpr.py", line 227, in main
bkrpr = cli.OcrGlob(opt)
File "/home/ken/Desktop/bkrpr-0.1.82/cli.py", line 88, in __init__
self.M = model.Model()
NameError: global name 'model' is not defined

[latchkeyed]

@kenbaldwin: Hi there! A couple replies for you.

Triggering: Yes, gphoto and gphoto2 are the right software for doing remote camera capture in linux. Unfortunately I was somewhat stymied trying to find cameras that are compatible and still currently available. But then, they also have multiple kind of supported cameras, from simply "capture" to "fully controllable". If you want to go the remote capture route, I'd love to hear more. I'm currently spending all my project-available funds on plastics and mount materials so I'm on hiatus from camera testing.

Command Line operation: Depends entirely on what version of the software you pulled down. If you've done a checkout recently you probably have our wxPython version. You can easily tell by whether or not you have page images inside the GUI. If you can see any pages, then you have the new version.

This is going to be our 0.2 release and is still currently in testing. At the moment I think the command line portion has been put on hold until 0.2 comes out, which will complete our transition from gtk to wxWidgets and also bring multi-platform support.

If you want to find out more, you can check out https://edge.launchpad.net/bkrpr

[kenbaldwin]

Thanks, latchkeyed.

I couldn't find any affordable cameras that supported remote capture either, so I'll probably go with the A590 for now. My local plastics shop is currently fabricating the 12" plexi piece for me for $35.

I downloaded the v0.1.82 tarball on your website. I haven't tried pulling the dev branch from bzr.

[kenbaldwin]

Where's the best place to discuss bkrpr hardware and software? There's no forum at bkrpr.org. Is this thread the only option right now?

[latchkeyed]

@kenbaldwin

Good question/point. I've created a google group at http://groups.google.com/group/bkrpr where we can get all the discussion in one place.