Kamscan

Kamscan implements a camera-based scan station for documents. It does so as a command line tool, making the process very convenient and efficient.

It captures pictures with a photo camera and converts them to a PDF document. Additionally, it leverages OCR software to embed the text as text in the PDF, making the PDF searchable and indexable.

Prerequisites

You need

• Unix-like OS
• Python 3.7
• Lensfun
• GCC
• Tesseract 4
• pdftk
• ImageMagick 6.8
• Argyll CMS (in particular, cctiff)
• dcraw
• Kivy
• pytz
• click (Python package)
• ruamel.yaml

Installation

1. Clone this repo.
2. make undistort

Note that you cannot simply make a link in your PATH pointing to kamscan.py. Instead, you have to write a trivial shell script named kamscan that calls it explicitly, like:

#!/bin/sh

PYTHONPATH=~/src python -m kamscan.kamscan "$@"

(In my case, kamscan installed in ~/src/kamscan.) This is because kamscan.py uses relative imports, and it looks for its utilities in the directory it was called in.

Then, you can call kamscan --help as a starting point.

Configuration

The configuration file must be placed at ~/.config/kamscan/configuration.yaml. It may look like this:

sources:
  sony_nex_7:
    make: Sony
    model: NEX-7
    icc_profile:
      path: /home/bronger/.config/darktable/color/in/nex7_matrix.icc
      color_space: Lab
    lenses: ["1", "2"]
    reuse_dir_prefix: /tmp/kamscan_nex_7
  sony_a6000:
    make: Sony
    model: A6000
    icc_profile:
      path: /home/bronger/.config/darktable/color/in/nex7_matrix.icc
      color_space: Lab
    lenses: ["1", "2"]
    reuse_dir_prefix: /tmp/kamscan_a6000
default_source: sony_a6000
lenses:
  "1":
    make: Sony
    model: E 50mm f/1.8 OSS (kamscan)
  "2":
    make: Sony
    model: E 50mm f/1.8 OSS

This configuration file contains two possible camera sources that you can select with the --source argument. The Sony α6000 is the default. Furthermore, two lenses are selectable. “Model” and “make” refer to the respective values in the Lensfun database.

Camera setup

The camera setup must be in a way that you take pictures similar to this:

,--image border----------.
|                        |
|  ,-paper border----.   |
|  |                  |  |
|  |   |  |  |  |  |  |  |
|  |   |  |  |  |  |  |  |
|  |   |  |  |  |  |  |  |
|  |                  |  |
|  x-----------------´   |
|                        |
`-----------------------´

The code that feeds the images into the core processing units rotates these images by 90° clockwise. This way, any text printed on the pages is readable without turning your head.

The setup need not be perfect. Any inhomogeneous illumination or tilting of the paper will be corrected.

Two-side setup

,--image border----------.
|                        |
|  ,-paper border----.   |
|  |        |        |   |
|  | -----  | -----  |   |
|  | -----  | -----  |   |
|  | -----  | -----  |   |
|  |        |        |   |
|  x--------+--------´   |
|                        |
`------------------------´

Usage

Some hints going beyond the output of kamscan.py --help:

• Profiles are stored only for one day. After that, Kamscan assumes that they are inaccurate.

• Switching between different page sizes, colour modes, or two-side mode does not need different profiles. Instead, create a new profile for a different setup, e.g.:

  • different illumination
  • a glass panel on top of the sheets
  • sheets in bigger distance because they are too large for the small one

  Of course, if you return to the former profile, you must be able to reproduce its setup accurately.

• Bear in mind that the height of the sheet used for creating the profile is assumed to be A4 by default. Use --full-height with other heights.
• The alignment point for everything you scan is always the top left corner (in reading orientation) of the rectangle used for calibration, marked with “x” in the above diagrams. Accordingly, the two edges of the calibration rectangle originating in this point are the alignment lines for everything you scan.

Limitations

Any bent pages are a problem. They should be as flat as possible because no correction of any curvature takes place.

Adding a new camera model

Place a new script in sources/. The name of the script (without the .py) must match the name of the key in the configuration file under “sources”.

The API that the class must fulfil is simple:

• name it “Source”
• accept the arguments “configuration” and “params” in the constructor
• define the methods “images” and “raw_to_pnm”

Constructor arguments

“configuration” is a nested dictionary with the part of the configuration file that belongs to the source.

“params” is the value that was passed with the --params argument on the command line. If no such argument was given, it is None. If it was only a single value, it is that value. If it was a comma-separated list of key=value pairs, it is a dictionary with those pairs.

The method “images”

This iterator yields the raw images from the camera in the order that they were taken. It yields a tuple of image index (starting with 0), whether it is the last page, and the path to the image. There must be at least one image in the iterator.

The method “raw_to_pnm”

This method converts a camera raw file into a PNM. It does so as raw as possible, i.e. no corrections. In particular, the colour space must be linear. It takes the following parameters:

path

The path to the raw file. The output file must have the same path but with the extension .ppm (colour) or .pgm (grey).

for_preview

(Default: False.) If True, generate a pnm that looks decently instead of being raw. In particular, it should be white-balanced and gamma-corrected.

gray

(Default: False.) Whether a PGM should be produced.

b

Exposure correction; all intensities are multiplied by this value. Default is no exposure correction.

asynchronous

(Default: False.) If True, the external process that does the conversion is called asynchronously.

The return type depends on the last parameter asynchronous. If it is False, the path to the PNM path is returned. Otherwise, a tuple is returned with the output path and the external process (of the type subprocess.Popen).