ML Model-Based Solution to Refine CAPTCHA

Problem Description

This project focuses on enhancing CAPTCHA systems using a machine learning (ML) approach. The goal is to improve security and user experience by:

• Dataset: Creating or using a dataset of CAPTCHA images.
• Model Training: Training an ML model to generate and validate CAPTCHA challenges.
• Security Features: Implementing adversarial techniques to counter automated CAPTCHA-solving bots.
• Deployment: Deploying a web-based CAPTCHA generator using Streamlit.

Project Overview

This project includes:

• Dataset Generation: Using the captcha library to generate synthetic CAPTCHA images.
• Data Preparation: Loading and preprocessing images and encoding labels.
• Model Training: Building and training a multi-output Convolutional Neural Network (CNN) to recognize CAPTCHA text.
• Adversarial Techniques: Incorporating methods (e.g., FGSM) to generate adversarial examples that challenge automated solvers.
• Deployment: A Streamlit app that generates CAPTCHA challenges, validates user input, and allows customization of CAPTCHA difficulty.

Project Structure

captcha_project/

├── captcha_dataset/ # Folder for generated CAPTCHA images

├── adversarial.py # Module for generating adversarial examples

├── app.py # Streamlit app for CAPTCHA generation and validation

├── data_preparation.py # Module to load and encode CAPTCHA images

├── generate_dataset.py # Script to generate the CAPTCHA dataset

├── requirements.txt # List of project dependencies

└── train_model.py # Script to build and train the CAPTCHA solver model

Installation

1. Clone the Repository:

   git clone https://github.com/karan3691/ML-Model-Based-Solution-to-Refine-CAPTCHA.git

   cd ML-Model-Based-Solution-to-Refine-CAPTCHA

2. Installation:

   python -m venv venv
   source venv/bin/activate   # On Windows: venv\Scripts\activate

3. Install Dependencies:

   pip install -r requirements.txt

Usage

1. Generate the CAPTCHA Dataset

   Run the script to create a dataset of CAPTCHA images:

   python generate_dataset.py

   This will generate CAPTCHA images and save them in the captcha_dataset folder.

2. Train the CAPTCHA Solver Model

   Train the ML model to recognize CAPTCHA images by running:

   python train_model.py

   The model is saved as captcha_solver_model.h5 after training.

3. (Optional) Adversarial Techniques

   The file adversarial.py contains a function to generate adversarial examples using FGSM. You can integrate this function into your CAPTCHA generation process to add extra noise that challenges bots.

4. Run the Streamlit Application

   Start the web application to generate and validate CAPTCHA challenges:

   streamlit run app.py

   The app includes:

   • input Section: Generates and displays CAPTCHA challenges.
   • Output Section: Validates user responses.
   • Enhancements: A slider to adjust CAPTCHA difficulty.

Model Hosting

Due to GitHub's file size restrictions, the trained model (captcha_solver_model.h5) is hosted on Hugging Face. To download the model within your code, use:

from huggingface_hub import hf_hub_download
model_path = hf_hub_download(repo_id="karan3691/captcha-model", filename="captcha_solver_model.h5")

Make sure your application references the model from the correct path.

Acknowledgements

• Captcha - Python library for generating CAPTCHA images.
• Streamlit - Framework for building interactive web apps
• Hugging Face - Platform for hosting and sharing ML models.