My hobby project for solving nonograms.
There is also a sub-project that recognizes nonograms.
The solver uses and the OCR produces nonogram descriptions in a simple JSON format, which liiks like this:
{
"row_hints": [
[1],
[1]
],
"col_hints": [
[1],
[1]
]
}
All the hints are arranged the way they are in the usual nonogram: left-to-right and top-to-bottom.
The solver is implemented in Rust. It uses branch and bound approach, together with an efficient line-by-line solving.
Ensure you have Rust and Cargo installed, and execute
cargo install --path .
You now can run it as nono-solver.
nono-solver [OPTIONS] [FNAME]
If FNAME is no specified, the input JSON will be read from stdin.
-m, --max-depth <MAX_DEPTH>: maximum depth of branching. 0 means no branching, solving solely line-by-line; it should be enough for "proper" nonograms. Specifying a number larger then necessary doesn't hurt, as the solver tries lesser depths first.-f, --find-all: if this flag isn't specified, the solver terminates when it finds the first correct solution. Specify the flag if you want to find all the solutions, or to check if the nonogram has only one solution.
The OCR is written in Python and uses OpenCV library. Use it to avoid manually typing the nonogram descriptions, if you want to just solve the puzzle you've found on a website.
Currently, ony the Katana website is supported.
Create and activate a Python virtualenv, if you want to. Then execute
pip install -e .
You can now run it as nono-ocr.
The solver takes an image from the first command-line argument, or, if unspecified, tries to read it from the clipboard. Take a screenshot from the Katana website and crop it just outsude the black border. Then feed to to the OCR. The nonogram description is produced to stdout; you can either save it so a file for later use, or feed it to the solver directly:
nono-ocr sample.png | nono-solver