Solving Handwritten Equation using Convolutional Neural Network
- OpenCV
- Keras
In this project I have tried to use opencv and pretrain resnet50 model to evaluate handwritten expressions. To test the project I have created handwritten expressions on paint and loaded the image into Evaluate_Equation.ipynb
1. Extract_data.ipynb
load images from dataset
image -> grayscale ->image negation
Find contours
sort by boundingRect
Find rectangle with maximum area
Crop image
Resize and reshape image to 1D Array
Append class ( as numbers 0 to 12 )
Store in list and convert to csv
2. Handwritten_train.ipynb
import csv using pandas
split into images and labels
convert 1D image to 3D image
Reshape image to (,28,28,3)
import pretrained Resnet50 model and add Dense layer
Train the model
Save the model
3. Evaluate_Equation.ipynb
import test image
Convert to grayscale and threshold
Find contours and sort it from left to right
Make list of boundingRect coordinates
loop through list and check if boxes are overlapping
Use boxes to crop image
Fit the model on the cropped image
Use eval function to evaluate expression
The Resnet model is trained for 10 epochs with a batch size of 200 to an accuracy of 98%
Test image :
Extracted Characters :
Evaluation Result :
