CCR

This is a library for recognizing handwritten Chinese characters + an SDL based frontend and an android app using it. Recognition is particularly tolerant to incorrect stroke order.

A conversion script for the KanjiVG database of Japanese characters is included. The recognition code should also work for other similar writing systems.

How to build

Build the tools, they don't require any libraries:

$ make -C tools simpl.elf makesc.elf

Make sure KanjiVG repo is cloned:

$ git submodule update --init data/kanjivg

Build ./kanji.list from KanjiVG.

$ ./data/kanjivg.py

./kanji.list should now exist. Build ./kanji.sc:

$ ./tools/makesc.elf < kanji.list > kanji.sc

For building the android app:

$ cp kanji.sc android/assets/kanji.list
$ cd android

Build the library for android:

$ ndk-build

Proceed with ant as usual, e.g:

$ ant debug

To build the SDL frontend:

$ make -C tools sdl.elf

On launch sdl.elf opens ./kanji.sc (in the current working dir). Compile sdl.c with -DDB_FILE=<other path> if desired. The SDL frontend requires SDL 1.x libraries, edit data/config.mk if needed.

Outline of the algorithm

Given a set of handwritten strokes, we compare it against all other sets in the database which have a similar number of strokes.

Given two sets of strokes (i.e. kanji), A and B, we construct a complete weighted bipartite graph (with A and B as parts) where the weight of an edge (a, b) is a measure of how dissimilar strokes a and b are . Then we find a minimum weight perfect matching (aka optimal assignment). The weight of this matching is how dissimilar A and B is.

TODO/Wishlist

• after a match, display a high quality drawing, not a simplified one
• display errors in stroke order

Files

• lib/

  • ccr.h
    • The recognition library.
  • klist_io.h
    • Utility for reading kanji.list
  • implementation(s) of parts of ccr.h, some unused.

• tools/

  • References ../lib
  • sdl.c
    • SDL 1.X based frontend, mouse and keyboard, outputs to stdout.
  • simpl.c
    • Used by ./data/build.sh to read svg curves from KanjiVG data.
  • makesc.c
    • Compiles klist to kanji.sc, which is used by the frontend.
  • Other misc tools.

• android/

  • Android frontend.
  • References ../lib

• kanji.list

  • A plaintext file containing simplified kanji strokes.
  • Can be read with lib/klist_io.h
  • Each kanji is a unicode codepoint and a list of strokes (in the correct order).
  • Each stroke is a piecewise linear curve described by a sequence of points.
  • Coordinates of all points are in [0,1].
  • Format:
    • for each kanji
      • <unicode codepoint (in decimal)>
      • for each stroke (in the proper stroke order)
        • let x = number of points
        • 2x numbers on a single line, separated by a single U+20, U+10 at the end

License

Copyright (C) 2018 Martin Shirokov

CCR is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

CCR is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with CCR. If not, see http://www.gnu.org/licenses/.