Multiple approaches to automated CAPTCHA solving using computer vision and deep learning. Built as a research project exploring different ML strategies for visual recognition challenges.

Automated browser interaction with TensorFlow-based CAPTCHA detection:

• captchacrack.py — Headless Chrome + TF frozen graph for real-time CAPTCHA solving
• captchacrack2.py — Improved version with better image preprocessing
• anatomia-captcha.py — CAPTCHA anatomy analysis and segmentation

Custom Convolutional Neural Network for character-level CAPTCHA recognition:

• captcha_model.py — CNN architecture definition
• captcha_train.py — Training pipeline with multi-GPU support
• captcha_eval.py — Model evaluation
• captcha_gen_default.py — Synthetic CAPTCHA generation for training data
• captcha_recognize.py — Inference script

Segment-then-classify approach:

• extract_single_letters_from_captchas.py — Image segmentation to isolate individual characters
• helpers.py — Preprocessing utilities (resize, threshold, contour detection)

┌─────────────┐     ┌──────────────────┐     ┌───────────────┐
│  CAPTCHA     │────▶│  Preprocessing   │────▶│  Recognition  │
│  Image       │     │  - Threshold     │     │  - CNN model  │
│              │     │  - Segmentation  │     │  - TF detect  │
└─────────────┘     │  - Contours      │     │  - Letter OCR │
                     └──────────────────┘     └───────────────┘

tensorflow>=1.13
opencv-python
selenium
beautifulsoup4
pillow
numpy

Disclaimer: This project was built for educational and research purposes to understand ML-based visual recognition. Use responsibly.

2019–2020