Shape-Detection-ML

A MATLAB-based shape detection system that classifies geometric shapes using both supervised learning and feature extraction methods (a bachelors course project 2020).

Project Overview

This project implements an intelligent shape detection system capable of classifying four different geometric shapes:

• Square (مربع - morabaa)
• Triangle (مثلث - mosalas)
• Circle (دایره)
• Hexagon (6 ضلعی - 6 zelii)

Features

• Dual Approach: Implements both supervised learning (Perceptron algorithm) and traditional feature extraction methods
• Bilingual Support: Persian text labels for shape classification results
• Interactive Testing: GUI-based file selection for testing custom images
• Robust Preprocessing: Advanced image processing pipeline including edge detection, noise removal, and morphological operations
• Real-time Classification: Immediate results display with shape identification

How It Works

Training Phase

1. Data Collection: Uses 20 training images (5 images per shape class) from the trainPic/ folder
2. Image Preprocessing:
   • Convert to grayscale
   • Edge detection using Canny algorithm
   • Noise removal and morphological operations
   • Resize to 50x50 pixels
   • Flatten to 2500-dimensional feature vectors
3. Model Training: Perceptron learning algorithm with binary encoding for each shape class

Testing Phase

1. User Input: Select any image file through the GUI
2. Processing: Apply the same preprocessing pipeline as training
3. Classification: Use trained model to classify the shape
4. Results: Display classification results with Persian text labels

Classification Encoding

• Square: [1 -1 -1 -1]
• Triangle: [-1 1 -1 -1]
• Circle: [-1 -1 1 -1]
• Hexagon: [-1 -1 -1 1]

File Structure

src code/
├── main.m              # Main execution file
├── getdata.m           # Training data preparation
├── getfeature.m        # Feature extraction implementation
├── removeBack.m        # Background removal utility
├── trainPic/           # Training images directory
│   ├── a1.jpg-a5.jpg   # Square training images
│   ├── b1.jpg-b5.jpg   # Triangle training images
│   ├── c1.jpg-c5.jpg   # Circle training images
│   └── d1.jpg-d6.jpg   # Hexagon training images
└── Untitled.m          # Additional utility file

Requirements

• MATLAB with Image Processing Toolbox
• Computer Vision Toolbox (for region properties)
• Statistics and Machine Learning Toolbox

Usage

1. Open MATLAB and navigate to the src code/ directory
2. Run main.m to start the shape detection system
3. The system will train on the provided images
4. Select a test image when prompted
5. View the classification results

Research Context

This project was developed as part of a "Research and presentation methods" course in a Bachelor's program (May 2020). It demonstrates the application of machine learning and computer vision techniques for pattern recognition and shape classification.

Technical Details

• Algorithm: Perceptron learning with binary classification
• Feature Dimension: 2500 (50x50 pixel images)
• Training Samples: 20 images (5 per class)
• Image Processing: Canny edge detection, morphological operations, noise removal
• Output: Binary classification with Persian text labels