Captcha Bypass Generation Using GANs
Developed GANs (Generative adversarial networks) to bypass and synthesize Captcha images, aiming to evaluate the robustness of Captcha security mechanisms.
Project Overview
This project leverages Generative Adversarial Networks (GANs) to generate realistic and complex captcha images, which are then used to train Convolutional Neural Networks (CNNs) for robust visual captcha recognition. The GAN is trained on a dataset of real captchas and is capable of producing new, diverse samples that improve the performance of downstream OCR models.
What It Does
- • Generates synthetic captcha images using GANs
- • Builds refined datasets for training and evaluating OCR models
- • Trains CNNs for accurate captcha recognition
- • Improves robustness against complex and adversarial captchas
Core Components
GAN Model
Trained to generate realistic captcha images that mimic real-world distributions.
CNN OCR
Convolutional Neural Network trained on both real and GAN-generated captchas for robust recognition.
Captcha Generator API
Live demo below: click Generate to see a new captcha image generated by the backend API.
Key Capabilities
- GAN-based synthetic captcha image generation
- CNN-based OCR for captcha recognition
- Dataset refinement and augmentation
Technologies Used
Other Projects
UMBC Interactive 4-Year Course Planner
Full-stack web app with Flask and React.js; Clerk auth and Firebase integration.
AI Fantasy World Generator
Text-based adventure game using the LIGHT dataset from FacebookAI and OpenAI's babbage-002 model for dynamic, AI-driven fantasy storytelling.
Subnet (Network Simulation)
Dockerized Python simulation for multi-subnet TCP/UDP testing.