🖼️ Content-Based Image Retrieval (CBIR)

A Content-Based Image Retrieval (CBIR) system implemented in C++ and OOP using OpenCV. The project demonstrates image search using classical visual features (SIFT, SURF, ORB, color histograms), similarity measurement, ranking, and a simple GUI demo.

🚀 Features

• Supports multiple feature extractors:
  • Color Histogram (HSV)
  • Color Correlogram
  • Histogram of Oriented Gradients (HOG)
  • SIFT
  • ORB
• Supports Bag of Visual Words (BoVW) for dimensionality reduction and compact image representation.
• Stores a forward index of image features in binary .bin files.
• GUI tools for:
  • Extracting features from a dataset
  • Querying an image and ranking results by similarity

🏗️ System architecture

1. Offline Phase (Indexing)

   • Extract features from images.
   • For SIFT / ORB / HOG → cluster descriptors with K-means → build BoVW histograms.
   • Save features and vocabulary into binary .bin files.

2. Online Phase (Query)

   • User selects a query image.
   • Extract features from the query image.
   • Compare with the stored index using:
     • Chi-square (for color features)
     • L2 / Euclidean (for local/BoVW features)
   • Sort and return Top-K results.

🧪 Experiments

• Datasets
  • CD dataset (99 train, 10 test)
  • TMBuD dataset (100 train, 35 test)
• Hardware: Intel i5-9600K, 64GB RAM
• Metrics: Extraction Time, Query Time, mAP (Mean Average Precision)

🔹 Key results

• SIFT + BoVW (500 visual words) achieved the highest mAP (~0.5) but with slow query times.
• ORB is the fastest and suitable for real-time scenarios, but with lower accuracy.
• HOG is stable but yields lower mAP.
• Color Histogram / Correlogram are fast to compute but less robust to illumination changes.

📸 Demo GUI

• Feature extraction UI and query interface that displays Top-K results.

⚙️ Development environment

• Visual Studio 2022
• C++17
• OpenCV 4.10.0

📂 Project structure

├── Content-based image retrieval/   # Source code
├── Release/                         # Executable
├── Docs/                            # Reports & documentation
├── Demo/                            # Video demo (use Git LFS if >100MB)
└── Dataset/                         # Datasets & generated feature database
    ├── CD_images/                   # 109 CD images
    └── TMBuD/                       # STANDARD subset of TMBuD (135 structure images)