hocr

Transcribe (OCR) image of hieroglyphic to Unicode

To run OCR on image of hieroglyphic text

These instructions were tested on Linux.

cd src
unzip gardiner.zip
unzip letters.zip
unzip newgardiner.zip
unzip topbibhiero.zip
python tables.py
python train.py
python evaltranscribe.py

Now look in directory transcriptions for the results.

This is experimental code at an early stage of development.

To run OCR on page possibly containing hieroglyphic text

Assume 1.png is an image of a page and 1.png.json is the output of Azure applied on that image, located in the same directory as 1.png.

python pipeline.py somepath/1.png

This will open a window for manually correcting recognition of hieroglyphic. Save it. Manually edit 1.png.csv in somepath to correct the hieroglyphic, e.g. using https://nederhof.github.io/hierojax/edit.html.

Run again:

python pipeline.py somepath/1.png

This will put HTML file 1.html in folder transcriptions.

To correct this manually, do:

cd transcriptions
correctpage.sh 1.html

Click on shaded areas to open a plaintext editor for these. Click again on a shaded area to close the editor (the result is not yet saved). One can toggle between views with or without the cutouts from the original scan and with or without colors to emphasize styles.

When done, press on the button at the top saying finish. This overwrites 1.html with the corrected version and will close the browser page.