This project implements a deep learning model to recognize both handwritten digits and characters using the EMNIST dataset. The EMNIST dataset is an extended version of the MNIST dataset, designed for both digit and character classification.
The EMNIST dataset consists of grayscale images of handwritten digits and letters, divided into multiple subsets (Balanced, ByClass, ByMerge, Digits, Letters, and MNIST). For this project, we use the EMNIST Balanced subset, which contains 47 classes (digits and letters).
You can learn more about the EMNIST dataset here.
The project uses a Convolutional Neural Network (CNN) for classifying the handwritten digits and characters. The CNN is designed to handle the complexity of multi-class classification with high accuracy.
The model achieves an accuracy of approximately XX% on the test dataset. Below are some visual representations of the results:
Here are a few samples from the EMNIST dataset:
Accuracy: 0.8914
Classification Report:
precision recall f1-score support
0 0.66 0.57 0.61 472
1 0.54 0.75 0.63 480
2 0.88 0.90 0.89 498
3 0.96 0.97 0.96 482
4 0.89 0.91 0.90 489
5 0.92 0.89 0.90 496
6 0.94 0.95 0.95 465
7 0.93 0.97 0.95 468
8 0.96 0.91 0.94 478
9 0.98 0.97 0.97 459
A 0.95 0.95 0.95 492
B 0.92 0.96 0.94 477
C 0.97 0.94 0.95 504
D 0.90 0.91 0.91 431
E 0.96 0.97 0.97 473
F 0.95 0.96 0.95 469
G 0.95 0.91 0.93 469
H 0.98 0.93 0.96 476
I 0.63 0.67 0.65 469
J 0.93 0.89 0.91 506
...
accuracy 0.89 17280
macro avg 0.89 0.89 0.89 17280
weighted avg 0.89 0.89 0.89 17280
The confusion matrix below illustrates the performance of the model across all classes:
