Handwritten Digit Classification Using CNN

Trained a CNN model on MNIST by applying image preprocessing, convolution-pooling feature extraction, dense classification layers, and deployed a custom prediction function for classifying new handwritten digit images.

🙃I am still working on this project please visit later for detailed view

Deep Learning & Frameworks

  • TensorFlow
  • Keras Sequential API

Model Architecture

  • Conv2D layers (32 & 64 filters)
  • MaxPooling layers (2×2)
  • Flatten layer
  • Dense hidden layer (128 units, ReLU)
  • Dense output layer (10 units, Softmax)

Data Preprocessing

  • MNIST dataset loading
  • Image inversion (255 - pixel values
  • Normalization (0–1)
  • Channel reshaping (28×28×1)
  • One-hot encoding of labels

Training Configuration

  • Optimizer: Adam
  • Loss: Categorical Crossentropy
  • Epochs: 10
  • Batch size: 200
  • Validation split: 20%

Deployment & Inference

  • Model saving: mnist_cnn.h5
  • Custom prediction pipeline using Pillow, OpenCV, NumPy
  • On-image confidence score visualization with Matplotlib