A Vision-Based Assistant for the Memory Game

Final project for the Image Analysis course (SSY098) at Chalmers University of Technology.

Overview

This project implements a computer vision system to assist a player in the card game "Memory". The system detects, classifies, and matches cards using computer vision and machine learning techniques, providing automatic support during the game.

Main Features

• Card detection using edge detection (Canny) and contour analysis
• Card classification (front/back) via:
  • SIFT feature matching
  • Convolutional Neural Network (CNN)
• Matching of face-up cards using SIFT feature matching
• Game logic for two players, with turn and score management
• Support for images from different perspectives and rotations

Project Structure

• main.ipynb: main notebook with the full processing, training, and testing workflow
• functions.py: support functions for feature extraction, matching, training, etc.
• card_classifier.pth: pre-trained CNN model for card classification
• images/: sample images, full games, training and test samples
• requirements.txt: Python dependencies
• report/: final report and related resources

Requirements

• Python 3.10+
• See requirements.txt for all dependencies (including OpenCV, torch, torchvision, matplotlib, scikit-learn, etc.)

To install dependencies:

pip install -r requirements.txt

Usage

1. Prepare images: place game images in the images/gameX/ folder (see examples).
2. Run the notebook main.ipynb in Jupyter or VS Code.
3. The system will automatically detect cards, classify front/back, and suggest moves.
4. Available functions:
   • Train the CNN classifier on new data (trainCardClassifier)
   • Generate new samples via augmentation (augmentDataset)
   • Test the system on single images or full games

Notes and Limitations

• Card detection may fail on very light backgrounds or when cards are touching.
• The system was tested on 1024x1024 px images, but can be adapted.
• Robustness is guaranteed for different rotations and perspectives, but depends on image quality.

Author

Gabriele Beltrami
Chalmers University of Technology
SSY098 - Image Analysis VT2025

References

• See report/ for the detailed report and experimental results.
• The project uses SIFT, CNN (PyTorch), OpenCV, and other open source libraries.