Convolutional Neural Network for Handwritten Digit Recognition

This project demonstrates the implementation of a Convolutional Neural Network (CNN) using Keras and TensorFlow for handwritten digit recognition on the MNIST dataset.

Table of Contents

• Introduction
• Requirements
• Usage
• Code Explanation
• Results
• Future Improvements
• License

Introduction

Handwritten digit recognition is a fundamental task in computer vision and has numerous applications in areas such as document processing, postal automation, and form data entry. This project aims to build a CNN model that can accurately classify handwritten digits from the MNIST dataset.

Requirements

• Python 3.x
• TensorFlow 2.x
• Keras
• NumPy
• scikit-learn

Usage

Clone the repository:

git clone https://github.com/your-username/handwritten-digit-recognition.git

Navigate to the project directory:

cd handwritten-digit-recognition

Run the script:

python digit_recognition.py

The script will train the CNN model on the MNIST dataset, evaluate its performance, and save the trained model to a file named ocr.h5.

Code Explanation

• The necessary packages are imported, including Keras layers, optimizers, and the MNIST dataset.
• The build_model function is defined to create the CNN architecture. It consists of two sets of convolutional, activation, and max-pooling layers, followed by two sets of fully connected, activation, and dropout layers. Finally, a softmax classifier is added.
• The MNIST dataset is loaded, and the data is reshaped and scaled to the range of .
• The labels are converted from integers to vectors using LabelBinarizer.
• The CNN model is compiled with the Adam optimizer and categorical cross-entropy loss.
• The model is trained using the fit method, with the training and validation data, batch size, and number of epochs specified.
• The trained model is evaluated on the test data, and the classification report is printed.
• The trained model is serialized to disk using model.save.

Results

The trained CNN model achieves an accuracy of approximately 99% on the MNIST test set. The classification report provides detailed performance metrics for each digit class.

Future Improvements

• Experiment with different CNN architectures and hyperparameters to further improve the model's performance.
• Implement data augmentation techniques to increase the diversity of the training data and improve generalization.
• Explore transfer learning by using pre-trained models as a starting point for fine-tuning on the MNIST dataset.

License

Lmao no license