CAPTCHA Solver

A Redis-integrated CAPTCHA solving system using TensorFlow and Keras.

Overview

This project allows you to:

• Train a model to solve CAPTCHA images with randomized alphanumeric text.
• Serve predictions through a Redis-based system using a lightweight Python listener.
• Preprocess and normalize images from a ./cropped directory.

Features

• Character set excludes ambiguous symbols like 0, 1, I, O, and l.
• Model architecture: CNN + BiLSTM + CTC decoding.
• Integrated with Redis for queue-based request/response processing.
• TensorFlow 2.x compatible.

Directory Structure

/cropped/            # Input CAPTCHA images, filenames are used as labels
trainer.py           # Trains the model and saves to captcha_model.h5
load.py              # Loads model, solves CAPTCHAs from Redis queue
requirements.txt     # Python dependencies
run.bash             # Optional startup script
index.php            # PHP interface to push CAPTCHA solving requests

Usage

Training

Place CAPTCHA images in the ./cropped folder with correct labels as filenames:

./cropped/8TrbK.png
./cropped/Yz7fU.png

Then run:

python3 trainer.py

Serving (Prediction)

Start the Redis server, then run:

python3 load.py

It will listen on the captcha Redis channel and write results using r.set().

Dependencies

Install Python packages:

pip install -r requirements.txt

License

MIT License

Author

Kaveh Sarkhanlou