A custom CNN-based CAPTCHA solver for image-based CAPTCHAs, built from scratch using Keras.

This project implements a convolutional neural network to automatically solve image-based CAPTCHAs. I designed and trained a simple custom CNN architecture.

Model Architecture:

• Custom CNN designed specifically for CAPTCHA character recognition
• Built using Keras/TensorFlow
• Trained on synthetically generated CAPTCHA dataset (captcha library)
• View the full data table here
• View the google sheets here (calculations)

Data Pipeline:

• CAPTCHA generation for creating training data
• Image preprocessing (grayscale conversion) to improve model performance

Key Features:

• Basic end-to-end solution from data generation to trained model
• Benchmarking results available in /results folder
• Simple, lightweight architecture focused on understanding functionality

• Framework: Keras/TensorFlow
• Image Processing: OpenCV (grayscale preprocessing)
• Dataset Generation: captcha

captcha-slayer/
├── results/          # Benchmark images and performance metrics
├── [model files]     # Trained model weights
├── [training code]   # Dataset generation and training pipeline
└── README.md

Performance benchmarks and accuracy metrics can be found in the /results folder, including visual demonstrations of the model's predictions.

• Designing CNN architectures from documentation
• Building basic data pipelines for ML training
• Simple image preprocessing techniques for improving model accuracy

• Expand support for different CAPTCHA types
• Optimize model architecture for better accuracy
• Clean up dependencies and document them

# 1. Create a virtual environment
python -m venv venv

# 2. Activate the environment
# On Windows: .\venv\Scripts\activate
# On macOS/Linux: source venv/bin/activate

# 3. Install requirements
pip install -r requirements.txt