Content-Based Image Retrieval (CBIR) System

Introduction

This repository contains a Content-Based Image Retrieval (CBIR) system that extracts features from a query image and retrieves similar images from a database. The system implements various image features, including color-based, texture-based, shape-based, and deep methods. It also supports feature fusion and dimension reduction.

Feature Extraction

Color-Based Features

• RGB Histogram: color.py

Texture-Based Features

• Gabor Filter: gabor.py

Shape-Based Features

• Daisy: daisy.py
• Edge Histogram: edge.py
• Histogram of Gradient (HOG): HOG.py

Deep Methods

• VGG Net: vggnet.py
• Residual Net: resnet.py

Local Descriptors

• Detected Corners Similarity using SIFT and Harris: local_description.py

Feature Fusion

Some features may lack robustness, and feature fusion is implemented for enhancement.

• Feature Fusion: fusion.py

Dimension Reduction

To address the curse of dimensionality, random projection is used.

• Random Projection: random_projection.py

Evaluation

The CBIR system evaluates images based on feature similarity using Mean Average Precision (MAP) metrics. The evaluation is implemented in evaluate.py.

Method	Color	Daisy	Edge	Gabor	HOG	VGG Net	ResNet
Mean MAP (depth=10)	0.614	0.468	0.301	0.346	0.450	0.914	0.944

How to Run the Code

Part 1: Building the Image Database

1. Clone the repository.
2. Create a directory named database and organize your images into subdirectories based on classes.

Part 2: Running the Code

• Streamlit User Interface

streamlit run gui.py

• Alternatively you can run search queries via query.py script.

Data Augmentation

Run the data augmentation script to get more images in the same directory.

python data_augmentation/augmentor.py <database_path> [flipv, fliph, noise, rot, trans, zoom, blur]

References

• Original Image Augmentation
• Original CBIR Project