baitcenter/koda

Deep Learning-enhanced Keyword Detection in Document Images

koda

Deep Learning-enhanced Keyword Detection in Document Images

koda was created by Matteo Pellegrino and Federico Terzi for the Computer Vision course held by Professor Di Stefano.

For an indepth explanation, check out the paper.

It consists of a pipeline to analyze document images and extracting/highlighting specific keywords, such as shown in this image:

A combination of Deep learning-based edge detection, Hough transforms, OCR and warping was used to achieve the result.

Usage

Start by building the docker image with (it may require using sudo on linux):

sudo ./build_docker.sh

Then you can either start the Flask server with:

sudo ./start_server.sh

Or you can open the Jupyter notebook with:

sudo ./start_jupyter.sh

And then navigate to http://localhost:8888