mnist_preprocessor

MNIST OCR image preproccesor program in Python

A GUI application for visualizing and preprocessing MNIST handwritten digit images with custom convolution kernels.

Features

• Display 10 sample images from each MNIST class (0-9)
• Refresh button to show new random samples
• Customizable preprocessing matrix with adjustable:
  • Kernel size (3x3 or 5x5)
  • Stride (1, 2, or 3)
  • Individual kernel element values
• Predefined kernel presets (Identity, Edge Detection, Blur, Sharpen)
• Side-by-side visualization of original and processed images

Requirements

• Python 3.6+
• PyTorch
• Torchvision
• NumPy
• Matplotlib
• Pillow
• Tkinter (included with most Python installations)

Installation

pip install torch torchvision numpy matplotlib pillow

Usage

Run the program with:

python main.py

The MNIST dataset will be automatically downloaded on first run.

How to Use

1. The application will initially display 10 random samples for each of the 10 digit classes (0-9)
2. Click "Refresh Samples" to get new random samples
3. Configure the convolution kernel:
   • Select kernel size (3x3 or 5x5)
   • Choose stride (1, 2, or 3)
   • Either input values manually in the kernel matrix or select a preset
4. Click "Apply Preprocessing" to see the processed images
5. Compare the original images (top) with the processed images (bottom).

Todo

• Multiple convolutions.