Awesome
Laser Book Scanning
DIY book scanning is about taking cheap consumer cameras and using them to quickly and cheaply scan books. A flatbed scanner is destructive and professionals use scanners that start at $10k and go up from there. The goal of this project is to get a high quality scanner that can be set up with only a couple of hundred dollars in equipment.
Thus far, DIY book scanners have typically reduced warping by physically forcing the pages to be flat using a V-shaped platen made of glass or plastic. Accomodating this platen complicates the design of the scanner greatly, makes scanning small volumes hard, and slows down the scanning process.
This project is about removing the need for a platen and using laser beams to detect the shape of the page for later dewarping. The laser-dewarp script handles two files for input. One is the normal well-lit page being scanned. The other is a laser shot taken of the same page moments before with the lights off and with laser lines turned on.
Requirements
To build a laser rig, you will need at least two line-focus lasers, a camera, good lighting, and suitable hardware to mount all of these things.
A Raspberry Pi is very useful to have as a way of automatically managing the pieces. The lasers and lights will need to turn on and off and the camera will need to take two pictures for every page scanned. The Pi can act as the manager of all of these systems.
After you have scanned some images, you will need the following in order to run the laser dewarping script:
- Python
- Numpy/Scipy
- OpenCV (python-opencv)
Usage
<pre> usage: laser-dewarp.py [-h] [--version] [--debug] [--output OUTPUT_PATH] [--upside-down] [--contrast CONTRAST] [--brightness BRIGHTNESS] [--greyscale] [--grayscale] [--deskew] [--laser-threshold LASER_THRESHOLD] [--stretch-factor STRETCH_FACTOR] input_path A program for dewarping images based on laser measurements taken during scanning. The input directory must have an image of just the background (background.jpg), just your hands over the background (hands.jpg), and the two horizontal lasers on the background (background-laser.jpg) and one or more scanned images. positional arguments: input_path Path where the documents to dewarp are stored. If this is a file, dewarp the single file. If this is a folder, dewarps all documents in the folder. The input directory must contain background.jpg, a hands.jpg, and a background-laser.jpg files and a "*-laser.jpg" file for every document to be dewarped. optional arguments: -h, --help show this help message and exit --version Get version information --debug Print extra debugging information and output pictures to ./tmp while processing. --output OUTPUT_PATH Path where the resulting dewarped documents are stored. Defaults to ./out --upside-down The source image is upside down, rotate 180 degrees before processing --contrast CONTRAST Adjust final image contrast (between 1.0 and 2.0) --brightness BRIGHTNESS Adjust final image brightness (negative means darker) --greyscale Output the resulting image in greyscale --grayscale Output the resulting image in grayscale --deskew Run a final content-based deskewing step before outputting. This analyzes the text itself. --laser-threshold LASER_THRESHOLD A threshold (0-255) for lasers when calculating warp. High means less reflected laser light will be counted. --stretch-factor STRETCH_FACTOR This parameter determines how much text will be stretched horizontally to remove foreshortening of the words. The stretching is concentrated near the spine of the book. When the lasers are far from the lens or the book is laid flat, it should be smaller. It is normally set between 1.0 and 5.0. </pre>Discussion
This is very early days and nothing is certain yet. To discuss ideas or ask questions, ask in the R&D section of the DIY Book Scanner Forum.
Acknowledgements
The main author of the code is Jonathon Duerig, though I have taken inspiration and ideas from Daniel Reetz, Christoph Nicolai (guitarguy) and anonymous2 from the diybookscanner.org forums.
Changelog
Version 1.0:
- Added automatic deskewing based on spine position
- Cut each photo into two scans based on the spine
- Add a final deskew operation based on content after the spine for fine-tuning.
- Add brightness, contrast, and greyscale conversion options
- Add support for upside-down cameras, an option which turns everything around 180 degrees
- Laser dewarp postprocessing is now feature complete.
Version 0.3/0.4:
- Add background and hand detection. Autocrop both hands and the background. Use the hands to determine the left/right edges of the content and page.
Version 0.2:
- Various tweaks to the algorithm to be more faithful to the original paper.
- Added --height-factor option which can be increased to reduce foreshortening artifacts.
Version 0.1:
- Improved spine detection.
- Added page edge detection
- Removed the rotate-and-crop preprocessing step. This was interfering with the dewarping.
- Dewarping now uses the spine and page edge detection to define the sides of the source shape.