Capchan – Solving CAPTCHA with Image Classification

This post documents a practitioner’s journey in defeating CAPTCHA challenges from a CTF, starting with deterministic pixel comparison for static images and progressing to a TensorFlow-based image-classification approach for distorted CAPTCHAs, ultimately packaged into a reusable CLI tool (capchan). It covers dataset collection from challenge feedback, model training/validation workflows, tuning and preprocessing choices, performance results, and practical limitations (notably grayscale constraints), emphasizing methodology and tooling rather than real-world incidents.