We're creating a calculator that recognizes handwritten math expressions and evaluate them.
It would be handy to have a calculator which evaluates handwritten math.
inside calculator directory run following command:
pip3 install -r requirements.txt
from root directory and run following command:
main.py [-h] [-t test image path] [-m model path] [--no-graphics]
optional arguments:
-h, --help show this help message and exit
-t [test image path] testing image on given path
-m [model path] set classifier model specified on given path
--no-graphics doesn't display graphics when running test file(s)
main.py
main.py -t samples/sample1.PNG
main.py -t samples --no-graphics -m recognition_model
We used CROHME 2014 dataset compiled by Harold Mouchère.
The dataset is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.
We first preprocessed dataset to become 100 x 100 images, and created label ranging from 0 to 21.
We used dilation method to thicken the line strokes, and added padding to make the square image before resizing to 100 x 100.
Then we splitted the datset into train set and test set (80% 20%).
To train our dataset, we built a Convolutional Neural Network(CNN) using tensorflow-keras.
The network architecture is inside calculator.network.models module.
This is the result of our training.
Hosung Lee
Andrew Hung
Spencer Bass
Harold Mouchère, ICFHR 2014 CROHME: Fourth International Competition on Recognition of Online Handwritten Mathematical Expressions (CROHME-2014) ,2,ID:CROHME-2014_2,URL:http://tc11.cvc.uab.es/datasets/CROHME-2014_2
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)
