Image Captcha Solver

Overview

Image Captcha Solver is a deep learning application built with PyTorch and served via a Django-based interface. It decodes captcha images containing lowercase letters and digits.

• CNN-RNN model for captcha recognition
• Django UI to upload and predict captcha images
• Continuous updates with new experiments and improvements

---

Features

• Accurate captcha recognition using CNN + GRU
• Simple Django interface for testing and visualization
• Integration-ready prediction class for backend or CLI usage
• Trained on synthetic captcha datasets with data augmentation.

---

Interface Preview

Repository Structure

Captcha-Solver/
├── captcha/
│ └── pyvenv.cfg, Scripts/
├── captcha_project/
│ ├── manage.py
│ ├── captcha_project/ (Django settings)
│ └── solver/ (app with views, models, forms, templates)
├── interface-images/ (sample UI screenshots)
├── model/ (trained models: best_model.pth, captcha_model.pth)
├── notebooks/ (training and prediction notebooks)
├── requirements.txt
└── Readme.md

Quickstart

1. Clone the repository:

git clone https://github.com/arushahmd/Captcha-Solver.git
cd Captcha-Solver

Create a virtual environment and install dependencies:

python -m venv captcha
source captcha/Scripts/activate  # Windows: captcha\Scripts\activate.bat
pip install -r requirements.txt

Run Django server:

python manage.py runserver

Open in browser: http://localhost:8000/ and upload a captcha image to test predictions.

Using Notebooks

• Open any notebook in notebooks/ for:
• Model training (training_notebook.ipynb)
• Testing predictions (prediction_notebook.ipynb)
• Experimenting with captcha reconstruction (reconstruct_experiment.ipynb)

Model Architecture

• CNN Layers for feature extraction
• Linear + Dropout
• Bidirectional GRU for sequence modeling
• Linear output for CTC decoding

Dataset

• Synthetic captcha images with lowercase letters and digits.
• Data augmentation applied via Albumentations for robustness.

Contribution

• Contributions are welcome. You can:
• Open issues for bugs or feature requests.
• Submit pull requests to add improvements.
• Follow the existing coding style and structure.

---

Author: Aroosh Ahmad — AI Engineer (NLP, LLMs, ML Systems)             GitHub • LinkedIn