This project implements a Convolutional Neural Network (CNN) to recognize handwritten digits with an impressive 98.84% accuracy. It includes a web-based interface for real-time digit recognition, allowing users to draw digits and receive instant predictions.
- Interactive web-based drawing interface
- Real-time digit recognition
- High accuracy (98.84%) on the MNIST dataset
- Responsive design for various devices
- Easy-to-use clear and predict functionality
- Python 3.8+
- TensorFlow 2.x
- Keras
- OpenCV
- NumPy
- Matplotlib
- HTML5 Canvas
- JavaScript
The CNN model consists of:
- 2 Convolutional layers with ReLU activation
- 2 MaxPooling layers
- Flatten layer
- Dropout layer (0.25 rate)
- Dense output layer with 10 units (one for each digit)
- OPtimized Dense layer rather than using softmax activation directly used optimized softmax for more accuracy
- Clone the repository:
git clone https://github.com/sohailshk/Digit-Recognition.git - Install the required packages:
pip install -r requirements.txt
3.run the noteobook on Google Collab
- Draw a digit on the canvas using mouse
- Click "Predict" to see the model's prediction
- Interactive Interface
The model was trained on the MNIST dataset using data augmentation techniques. To retrain the model:
- Ensure you have the MNIST dataset
- Run the training script:
python train_model.py
- Accuracy on test set: 98.7%
- Training time: Approximately 2 minutes on Google Colab GPU
##SCREENSHOT
Contributions to this project are welcome! Please fork the repository and submit a pull request with your changes.
This project is licensed under the MIT License - see the LICENSE file for details.
- The MNIST dataset providers
- TensorFlow and Keras documentation
- OpenCV community
For any queries, please reach out to [Sohail] at [sohailsaif504@gmail.com].