Scanning Books

Random notes for now until this becomes more important.

Most dewarpers assume a linear or quadratic dewarping. However, reality is more complicated…

• If you don't intend to regenerate a book, then you only need to capture the page edges exactly! All we need is the page text … and formatting … and pictures.
• Booksorber might be good enough for DIY. Booksorber results. Seems to use edge of book to do dewarping, a nice trick. Why didn't he use the lines in the text though??!
• There are some more results using laser scanning.

Spreads automates the entire process (including tedious research) quite well. Picks all the right programs and is super-easy in Ubuntu. Optional packages?? zlib1g-dev libpng12-dev libtiff4-dev libjpeg62-dev libxrender-dev

sudo apt-get install ruby-dev libmagickwand-dev libboost-all-dev 
sudo gem install pdfbeads

Really nice smartphone paper (blurring, adaptive binarization, dewarping?). Another nice paper, references their previous paper.

Following OpenCV tutorial, and used this chessboard at 110% zoom on my Thinkpad.

• Maybe add a handle. Maybe not…
• 
  • A way to screw in a camera using bolt and wingnut. Also, reinforced design for platen.

Cheap/free Instructable. Use a cardboard box, old picture frames (glass/plastic), and a camera with tripod!

Great test that generates test text and sees how several perform.

OpenOCR is a great service combining Google's Tesseract OCR and Docker, which I want to learn about more.

To install Tesseract from source

'tesseract myscan.png -l eng out', more at readme

Optimal letter height in pixels (less pixel scanning, other operations) is … 40 pixels? Also, black and white probably speeds it up. You can also set a specific dictionary or train a “font” of characters.

Leptonica does do dewarping, just fine automagically too!

• He is able to correct vertical disparity by drawing straight lines through text, but needs to normalize text spacing to figure out horizontal disparity. Hmmm…test.
• But Tesseract doesn't use it?!?! Although it seems to use leptonica for other stuff…Maybe do a test where you warp images and see what happens, maybe plot with hOCR data too.
• OpenCV use SO post 1 and OpenCV Remap (usage). Apparently speed is .03 sec with Mathematica or Intel Pperformance Primitives.

Bad experience with Tesseract?

• Can the software do the whole process with only 1 camera? Or do I need two…
• How does ScanTailor do image warping? Why can't it handle non-flat pages? The algorithm explanation looks good…
• Merge OCR back into image? Not for this app, but maybe future. Use pdftk. 1, 2

Goose is one standalone but bulky option. Maybe better is Readability API

USB shutter (toggling the 5V line on and off, or just plug and unplug usb from a usb port on computer) can work with CHDK, not sure of other cameras.

Lots of cameras have PTP protocol built-in for transfer of files over usb and some for even remote control (shutter, preview, etc). Gphoto remote control has a great list. For Canon cameras,CHDK's PTP addons enables some more flexibility than Gphoto.

• Dad's Nikon D90 allows remote preview out of box apparently with gPhoto

• Note: Gphoto states that Canon Powershot cameras after 2009 do not have remote control (shutter, preview, etc) enabled anymore, even if you use CHDK. (my $50 camera >.<)

I have a Canon Powershot A3000, but only browse and download of files is enabled.

scan em