How luvk1412/sign-language-to-text works
This repository is a webcam-based American Sign Language fingerspelling-to-text application, with accompanying data collection, image preparation, and CNN training scripts. The main user workflow captures hand images, preprocesses frames, predicts letters, assembles words and sentences, and offers Hunspell spelling suggestions through a Tkinter interface. Training and preprocessing are distinct supporting stages. Runtime interactions are based on the shown excerpts and README; the excerpts do not establish how collected data feeds preprocessing, and the README does not describe the additional specialized models.
Main entry points: app.py Live application · preprocessing.py Dataset preprocessor · collect-data.py Image collector · train.py CNN trainer
Components 16
Dataset preparation
- Image collector
collect-data.py - Captured images
- Dataset preprocessor
preprocessing.py - Image thresholding
image_processing.py - Training data
Model training
- CNN trainer
train.py - CNN model files
model/model-bw.h5
Live translation
- Live application
app.py - Frame preprocessing
app.py - Letter prediction
app.py - Word and sentence
app.py - Translation display
app.py
Text assistance
- Spelling suggestions
app.py
Other components
- Signer
- Webcam
- Hunspell dictionary
Connections 19
- Signer → Webcam: signs to
README.md - Webcam → Image collector: provides frames
collect-data.py - Image collector → Captured images: saves images
collect-data.py - Captured images → Dataset preprocessor: reads image folders
- Dataset preprocessor → Image thresholding: calls
preprocessing.py - Dataset preprocessor → Training data: writes images
preprocessing.py - Training data → CNN trainer: loads directories
train.py - CNN trainer → CNN model files: saves weights
train.py - Signer → Live application: uses
README.md - Webcam → Live application: captures frames
app.py - Live application → Frame preprocessing: processes frames
app.py - Frame preprocessing → Letter prediction: passes threshold image
app.py - CNN model files → Live application: loads weights
app.py - Live application → Letter prediction: calls prediction
app.py - Letter prediction → Word and sentence: updates letters
app.py - Live application → Spelling suggestions: requests suggestions
app.py - Spelling suggestions → Hunspell dictionary: queries dictionary
app.py - Live application → Translation display: updates interface
app.py - Word and sentence → Translation display: shows text
app.py