Apurba Bangla Optical Character Recognition (OCR)

Implementation and Results

Convolutional Neural Network

For NumtaDB Dataset our overall Validation Accuracy is 98.24%.


Confusion matrix for CNN on NumtaDB

Results of the samples from the NumtaDB

For BDRW Dataset our overall Validation Accuracy is 91.87%.


Confusion matrix for CNN on BDRW dataset.

Results of the samples from the BDRW

Stochastic Gradient Descent

For NumtaDB Dataset our overall Validation Accuracy is 92.83%.


Confusion matrix for SGD on NumtaDB

For BDRW Dataset our overall Validation Accuracy is 84.68%.


Confusion matrix for SGD on BDRW dataset.

Multi-layer Perceptron

For NumtaDB Dataset our overall Validation Accuracy is 84.49%.


Confusion matrix for MLP on NumtaDB

For BDRW Dataset our overall Validation Accuracy is 65.00%.


Confusion matrix for MLP on BDRW dataset.

K-Nearest Neighbor (KNN)

For NumtaDB Dataset our overall Validation Accuracy is 67.32%.


Confusion matrix for KNN on NumtaDB

For BDRW Dataset our overall Validation Accuracy is 77.06%.


Confusion matrix for KNN on BDRW dataset.

Random Forest Classifier

For NumtaDB Dataset our overall Validation Accuracy is 73.50%.


Confusion matrix for Random Forest on NumtaDB

For BDRW Dataset our overall Validation Accuracy is 56.66%.


Confusion matrix for Random Forest on BDRW dataset.

Overall Results

Overall Result