CAPTCHA Recognition Project

This project focuses on developing a robust CAPTCHA recognition system using deep learning techniques. Two approaches were implemented: a single model approach that processes the entire CAPTCHA image and a multiple classifiers approach, where separate classifiers are trained for each character position. Both methods utilize a ResNet50 architecture. A Streamlit application integrated with FastAPI was also developed to provide an interactive demonstration of the system's capabilities. The project underscores the effectiveness of deep learning in CAPTCHA recognition and provides a flexible, scalable solution for text-based CAPTCHAs.

---

Features

1. Two Recognition Approaches:

• One-go Recognition: Recognizes the entire CAPTCHA image in one step.

• Sequential Recognition: Divides the CAPTCHA into individual characters and uses five classifiers to recognize each character sequentially.

2. Frontend and Backend:

• Streamlit Frontend: Provides an interface to upload CAPTCHA images and view recognition results.

• FastAPI Backend: Handles predictions and model interaction.

3. Custom Dataset Training:

• Users can train the models on their own datasets by configuring paths, CAPTCHA length, and alphabet.

4. Preprocessing Tools:

• A script (preprocess.py) to clean the dataset by removing corrupted files.

---

Installation and Setup

1. Clone the repository:

git clone https://github.com/EninDmitriy96/CAPTCHA_recognition
cd <path_to_the_project>

2. Install required dependencies:

pip install -r requirements.txt

3. Ensure your dataset is prepared in the proper format:

CAPTCHA filenames should correspond to their decryption (e.g., abc12.png for CAPTCHA abc12).

Place CAPTCHA images in the data/ directory or configure paths for your dataset.

Usage

Running the Application

1. Start both the frontend and backend:

python run_all.py

2. Access the applications:

• Frontend: http://127.0.0.1:8501

• Backend: http://127.0.0.1:8000

3. Terminate both processes by pressing Enter in the terminal.

---

Using the Frontend

1. Upload a CAPTCHA image:

• Use examples from the data/ folder for best results (same dataset used for training).

2. View Predictions:

• The results from both approaches (one-go and sequential recognition) are displayed.

---

Training on Your Dataset

1. Update the following parameters in the code:

• Dataset Paths: Ensure paths to your dataset are correct.

• Sequential Classifiers: Set the number of classifiers according to the CAPTCHA length.

• Alphabet: Define the alphabet used in your CAPTCHAs.

2. Run the training script (code/models/notebooks):

---

Preprocessing the Dataset

To remove corrupted files from your dataset, run:

python code/datasets/preprocess.py

Future Improvements

• Add support for more CAPTCHA formats.

• Improve accuracy for custom datasets.

• Extend frontend for additional functionalities (e.g., batch processing).

Contributing

Contributions are welcome! Please open an issue or submit a pull request for any improvements or bug fixes.

---