edwardclem / santak-utils

Data (pre)processing for cuneiform character recognition.

santak-utils

Data (pre)processing for cuneiform character recognition.

Dependencies

The scripts in this respository require:

• numpy
• tqdm
• OpenCV 3
• h5py
• Pillow

Generating Character Images

render_chars.py renders all characters in a unicode code point range from the provided ttf font file. The rendered characters are saved in PDF format to an output folder.

To use, run:

render_chars.py [-h] [--code_range CODE_RANGE CODE_RANGE] [--font FONT] [--fsize FSIZE] [--outf OUTF]

Generating HDF Data Archives

gen_hdf.py performs edge detection on all images in a user-supplied folder, and reduces the corresponding contours via subsampling. Subsampling is performed by skipping edge points. Use of HDF5 features is limited due to OpenCV's limited HDF5 API. The unicode code points of each character are stored in a dataset in the root group named "unicode".

The source images and output contours are saved in datasets with groups "/imgs", "/contours" - dataset names are integer unicode code points. Text descriptions are found in dataset "/descs" with indices corresponding to the characters in "unicode".

To use, run: gen_hdf.py [-h] [--imgs IMGS] [--skip SKIP] [--min MIN] [--out OUT]

#Example

pipeline.sh contains example usage.