CV Visual Search Engine & Classifier

---

📌 Project Overview

This project implements a complete Content-Based Image Retrieval (CBIR) and Image Classification system using the MSRC Object Category Image Database v2 (591 images across 20 object classes).

The goal is to investigate which computer vision descriptors best capture visual similarity in a dataset with high inter-class ambiguity (e.g., grass vs. cows on grass). The system evaluates:

• Global Color Histograms
• Spatial Grid-based Color & Texture
• Texture Descriptors (LBP, EOH)
• Local Features with SIFT + Bag of Visual Words (BoVW)

🔑 Key Result

A Spatial Grid-based Local Binary Pattern (LBP) descriptor
(4×4 grid, P=10, reduced using PCA) outperformed more complex BoVW models:

• mAP: 0.158
• SVM Classification Accuracy: 68.07%

---

🛠️ Implemented Descriptors

Descriptor	Visualization
Global Color Histogram
Edge Orientation Histogram (EOH)
Local Binary Patterns (LBP)
SIFT + BoVW
Spatial Grid (Color + Texture)

---

📁 Repository Structure

CV-Visual-Search-Engine/
│
├── data/
│   └── MSRC_ObjCategImageDatabase_v2/
│       ├── Images/                    # Place all .bmp images here
│       └── descriptors/               # Generated .mat files appear here
│
├── src/
│   ├── config.py                      # ⚠️ MAIN FILE TO CONFIGURE
│   └── ...                            # Other Python scripts
│
├── assets/
│   ├── search.png
│   └── descriptor_visuals/
│       ├── EOH.png
│       ├── hist.png
│       ├── lbp.png
│       ├── SIFT.png
│       └── sp_grid.png
│
├── report_visuals/                    # Auto-generated results (PR curves, PDFs)
└── requirements.txt

---

⚙️ Configuration (IMPORTANT)

All paths are centrally managed via:

src/config.py

The project automatically detects its root directory, so it works regardless of where the repository is cloned.

---

🚀 Usage Guide

1️⃣ (Optional) Bag of Visual Words (BoVW Pipeline)

If you want to use SIFT + BoVW, you must generate the codebook first:

python src/bovw_step1_extract_all_sift.py
python src/bovw_step2_create_codebook.py

2️⃣ Feature Extraction

Generate descriptors for the dataset.

Edit src/config.py to select the descriptor mode (e.g., lbp, global, grid_col_tex):

python src/cvpr_computedescriptors.py

3️⃣ PCA & Retrieval Evaluation

Reduce dimensionality and compute retrieval metrics:

python src/compute_pca.py
python src/compute_evaluation_stats.py

Outputs:

• PCA scree plots
• Precision–Recall curves
• Mean Average Precision (mAP)

4️⃣ Automated Batch Experiments

Reproduce all experiments used in the report:

python src/run_experiment_batch.py

This script sequentially runs:

• Descriptor extraction
• PCA
• Retrieval evaluation

5️⃣ SVM Classification

Train an SVM using the best-performing descriptor (LBP + PCA):

python src/SVM.py

Outputs:

• Confusion matrix
• Per-class accuracy
• Overall accuracy (68.07%)

6️⃣ Visual Search Demo

Run an interactive visual search query:

python src/cvpr_visualsearch.py

---

📊 Results Summary

Descriptor Strategy	Best Metric	mAP	P@15
Spatial LBP (4×4, P=10, PCA)	L2	0.158	0.397
Spatial LBP + Color (PCA)	Mahalanobis	0.142	0.370
Spatial Grid (Color + Texture)	L3	0.120	0.326
Global Color Histogram	L1	0.096	0.267
SIFT + BoVW (k=500, PCA)	L2	0.083	0.254

---

📦 Requirements

Install dependencies with:

pip install -r requirements.txt

requirements.txt

numpy
scipy
opencv-python
matplotlib
scikit-learn
scikit-image
pandas
seaborn
joblib

---

📝 License

This project is available under the MIT License.

🤝 Contributing

Contributions, issues, and feature requests are welcome!

📧 Contact

For questions or feedback, please open an issue in this repository. Or contact me directly via email: tahir.hasan.thk@gmail.com