A web-based editor for Tesseract box files.

A box file contains lines with individual characters and their bounding boxes:

t 700 1000 750 1050 0
h 750 1000 800 1050 0
e 800 1000 850 1050 0

If you want to train Google's Tesseract OCR tool, you'll need to work with box files.

When run in "training mode", Tesseract outputs box data rather than plain letters. You correct the errors in these box files and feed them back into Tesseract. boxedit helps you do that.

======= Try a Live Demo!

Usage:

1. Open index.html in your favorite browser.

2. Drag a .box file and a corresponding image file onto the page. You can drag both at the same time or drag them one by one:

3. Fix errors by: 1. Editing the raw box data in the text box on the left. The boxes will update on the right. This is handy for deleting boxes or scootching them around. 2. Clicking a box in the image to select it and typing a letter. This will change the corresponding box data and advance the selection, allowing you to type the text that you see. 3. Clicking a box and using the "split" menu to fix merged characters.

To get going, run:

git clone https://github.com/danvk/boxedit.git
cd boxedit
npm install
python -m SimpleHTTPServer

and then open localhost:8000 in your browser of choice. You'll have a standard edit/save/reload iteration cycle.

To update the demo:

mkdir demo
jsx --harmony box.js > demo/box.js
jsx --harmony index.js > demo/index.js
cp style.css demo/
cp index.html demo/
cp index.html demo/demo.html
cp node_modules/react/dist/react-with-addons.min.js demo/

rm -rf ../danvk.github.io/boxedit
cp -r demo ../danvk.github.io/boxedit

Then modify demo.html to set 'demo mode' (second param to render).