Sign Language MNIST Dataset

The Sign Language MNIST dataset is a real-world-inspired, drop-in replacement for the classic MNIST dataset. It presents a challenging multi-class problem with 24 classes of American Sign Language (ASL) hand gestures representing the alphabetic letters (A-Z), excluding J and Z, which require motion.

Dataset Overview

• Image Format: Grayscale, 28x28 pixels
• Labels: Numeric (0-25) mapped to letters A-Z
• Training Data: 27,455 samples
• Test Data: 7,172 samples
• Format: CSV with rows containing labels and 784 pixel values

Data Creation

The dataset was created using an extensive pipeline:

1. Original color images were cropped to hands-only, gray-scaled, resized, and augmented.
2. Augmentations included:
   • Filters (Mitchell, Robidoux, Catrom, Spline, Hermite)
   • Brightness/contrast adjustments (+/-15%)
   • 3-degree rotation
   • 5% random pixelation

These enhancements improved the dataset's resolution and class separation, offering a unique challenge for machine learning models like Convolutional Neural Networks (CNNs).

Applications

A robust visual recognition model trained on this dataset could:

• Benchmark modern computer vision models
• Enable ASL translation tools for the deaf and hard-of-hearing
  Potential integrations include affordable devices like Raspberry Pi with OpenCV and Text-to-Speech capabilities.