Given images of words, recognize the words and output the corresponding text.
To use the docker image, first pull the image using
docker pull jumpst3r/wordimage2text
And then execute
docker run -it --rm -v /PATH_TO_FOLDER_WITH_INPUTS:/input/figures -v /FULL_PATH_TO_OUTPUT_FOLDER/:/output/ jumpst3r/wordimage2text sh /input/script.sh /input/figures/example.png /input/figures/boxes.csv /output/
where PATH_TO_FOLDER_WITH_INPUTS is the full path to a folder containing the inputs described bellow_
The input consists of:
example.png: A document image containing possibly several words.boxes.csv: A csv file containing the bounding boxes delimiting each word in the previously provided image
The output consists of:
-
A csv containing the coordinates of the words with the appended detected word:
x1,x2,y1,y2,detected_word -
A zip file containing the cropped out words and a csv which can be used to trace back the original words positions and textual content.
The docker image is also compatible with DIVAServices a web-based framework providing streamlined access to DOI methods.
Original repo