Chess Board Detection and Best Move Prediction using Machine Learning

Overview

This project implements a simpler machine learning-based system for identifying chess pieces from an image of a chessboard, generating the corresponding FEN (Forsyth-Edwards Notation), and determining the best move using a chess engine. The primary focus lies in recognizing individual pieces on the board, constructing a board array for the Python Chess library, and leveraging it to compute the best move.

Features

• Chess Piece Detection: Uses a trained machine learning model to identify individual pieces (both white and black) on the board.
• FEN Generation: Converts the identified board state into a valid FEN string.
• Best Move Prediction: Integrates with a chess engine to compute the best move based on the generated FEN.
• Web Application: A user-friendly web interface where users can upload chessboard images, specify whose turn it is (white or black), and receive the best move.
• Customizable Dataset: Train the model on other chessboard formats using the provided code and dataset generation tools.

---

Project Structure

1. Folder: Dataset and Machine Learning

This folder contains all code and resources related to creating the dataset, training the model, and making predictions:

• Dataset Creation: Scripts for generating datasets and CSV headers from chessboard images.
• Model Training: Code for training a machine learning model using the provided dataset.
• Board Prediction: Scripts to predict the board state using the trained model.
• Best Move Prediction: Logic to determine the best move using the predicted board state and a chess engine.

Dataset Details

• Source: The dataset was created using screenshots of chess board taken from chess.com.
• Statistics:
  • 120 board images were used for training.
  • Extracted 3,439 white piece images and 3,437 black piece images.
• Customization: You can expand the dataset by adding more images of chess pieces and retrain the model using the provided scripts.

2. Folder: Trained Model and Web Application

This folder contains the web application built using Flask and Python:

• Web Interface: Upload an image of the chessboard, specify whose turn it is (white or black), and get the best move.
• Chess Engine Integration: Requires the user to specify the path to the chess engine executable.
• Requirements: Includes a requirements.txt file to simplify dependency installation.

---

Installation and Setup

Prerequisites

1. Python 3.8+
2. NumPy
3. Pandas
4. Flask
5. OpenCV
6. Scikit-learn
7. Pillow
8. A compatible chess engine (e.g., Stockfish)

Steps to Run the Web App

1. Clone the repository:

   git clone https://github.com/LovejeetM/Chess_ML_Algorithm.git
   cd Chess_ML_Algorithm

2. Navigate to the Trained Model Webapp folder:

   cd "Trained Model Webapp"

3. Install dependencies:

   pip install -r requirements.txt

4. Set up the chess engine path in the web application:

   • Edit the ml.py file (or the appropriate script) to include the path to your chess engine executable.

1. Run the Flask application:

   python app.py

2. Open the web app in your browser at http://127.0.0.1:5000.

Additionally, you can use virtual environment if it does not run on your system and for testing.

---

Usage

Web Application

1. Upload Image: Upload a chessboard image (e.g., a screenshot from chess.com).
2. Specify Turn: Indicate whether it is White's or Black's turn.
3. Generate Best Move: The app processes the image, generates the FEN, and displays the best move.

Training a Custom Model

1. Add new chessboard images to the dataset.
2. Use the scripts in the Dataset and Machine Learning folder to generate piece images and prepare a CSV file.
3. Retrain the model using the provided training scripts.
4. Update the model file in the web app folder to use the new model.

---

Accuracy

• Current Model Accuracy: Achieves 100% accuracy on the provided dataset of chess board images.
• Custom Datasets: Accuracy may vary based on the quality and diversity of the training dataset.
• Light Weight Model: This codebase uses simple method for classification of pieces board.

---

Future Work

• Support for Other Chessboard Styles: Expand the dataset to include different board formats.
• Real-Time Detection: Implement live video feed detection for real-time chess analysis.
• Enhanced Web Features: Add more functionality to the web app, such as game history tracking and multi-move prediction.

---

License

This project is licensed under a custom license. Usage of this code is restricted to non-commercial purposes only. Modifications are allowed, but commercial redistribution is prohibited without permission.

---

Feel free to Connect

LinkedIn: https://www.linkedin.com/in/lovejeet-singh-matharu-975679213/ 

Feel free to ask any questions too.