mle8-capstone

American Sign Language (ASL) Classifier using Convolution Neural Networks (CNN)

By: Sheilah K. (@skirui-source)

Metadata

• The Sign Language MNIST data sourced from Kaggle posted by unknown author with username "tecperson" ---- URL : https://www.kaggle.com/datamunge/sign-language-mnist

• The data stored in .csv format represents American Sign Language hand gestures; 28x28 grayscale images as rows of labels 0-25 (first column) and pixel values 1-784 (remaining columns).

• Total 24 labels represent 1:1 mapping for letters A-Z, excluding cases for 9=J and 25=Z which require hand gesture motion.

• training and test data contain 27,455 and 7172 images respectively.

Data Lineage

• Original: color images of multiple users repeating the gesture against different backgrounds.

• To create new data, an image pipeline was used based on ImageMagick:

  • cropping to hands-only
  • gray-scaling, resizing,
  • creating at least 50+ variations to enlarge the quantity.
  • modification and expansion strategy: filters ('Mitchell', 'Robidoux', 'Catrom', 'Spline', 'Hermite')
  • 5% random pixelation
  • +/- 15% brightness/contrast
  • 3 degrees rotation