ASL Image Classification App

Problem Statement

Recognizing American Sign Language (ASL) hand signs from images is a challenging computer vision task. This project aims to build an interactive web application that can classify ASL alphabet signs (A-Z, space, nothing, del) from uploaded or sample images.

Solution

• A custom Convolutional Neural Network (CNN) was trained on the ASL Alphabet dataset.
• The model achieves ~99% accuracy on the validation set.
• The app is built with Streamlit and allows users to:
  • Upload their own ASL hand sign images.
  • Select sample images from the sidebar.
  • View the predicted letter and a probability bar graph for all classes.

Model

• The model is a custom CNN (see Notebook/ASL_Classification.ipynb for architecture and training).
• Trained on 64x64 RGB images.
• Achieves high accuracy (~99%) on validation data.

How to Use

1. Clone or download this repository.
2. Install dependencies:

   pip install streamlit tensorflow pillow matplotlib numpy
   

3. Add your trained model:
   • Place your Keras model file (my_model.keras) in the Model folder.
4. Add sample images:
   • Place sample images named A.jpg, B.jpg, ..., space.jpg, nothing.jpg, del.jpg in the Images folder.
5. Run the app:

   streamlit run app.py
   

6. Open the app in your browser and either upload an image or select a sample from the sidebar.

Project Structure

ASL_Classification/
│
├── app.py
├── Model/
│   └── my_model.keras
├── Images/
│   ├── A.jpg
│   ├── B.jpg
│   └── ... (other sample images)
├── README.md
├── .gitignore
└── Notebook/
    └── ASL_Classification.ipynb