CBIR-Code-

Causality-enhanced multiresolution residual learning framework for image retrieval. Combines causal feature refinement with hierarchical representations and Fast Osprey Optimization to improve robustness, efficiency, and retrieval accuracy across diverse datasets.

CBIR Causality-Enhanced Framework

This repository contains the implementation of the paper:

“A Causality-Enhanced Multiresolution Residual Learning Framework for Image Retrieval with Fast Osprey Optimization”

Authors

Abdulrahman Yousif Zeain, Abdullahi Abdu Ibrahim
Altınbaş University, Istanbul, Turkey

Overview

This work proposes a novel content-based image retrieval (CBIR) framework that integrates:

• Multiresolution feature extraction using ResNet-50
• Causal Variational Autoencoder (CausalVAE) for latent representation learning
• Fast Osprey Optimization Algorithm (FOOA) for feature selection

The framework enhances semantic consistency, reduces redundancy, and improves retrieval accuracy and efficiency.

Methodology

The proposed pipeline consists of:

1. Multiscale feature extraction using ResNet-50
2. Latent representation learning via CausalVAE
3. Feature optimization using Fast Osprey Optimization
4. Similarity computation using:
   • Structural Similarity Index (SSIM)
   • Histogram Intersection

Datasets

The experiments are conducted on publicly available datasets:

• CIFAR-10
• Oxford Flowers (102 categories)
• Corel-1000

Requirements

• Python 3.8+
• TensorFlow 2.8
• NumPy
• Pandas
• Scikit-learn

Install dependencies: python main.py

Experimental Setup

• Ubuntu 20.04 LTS
• NVIDIA GPU (≥12 GB)
• 64 GB RAM

Notes

• This repository provides a clean and reproducible implementation of the core framework.
• Large intermediate files and additional experimental outputs are available upon reasonable request.

Citation

If you use this work, please cite:

Zeain, A.Y., Ibrahim, A.A.
“A Causality-Enhanced Multiresolution Residual Learning Framework for Image Retrieval with Fast Osprey Optimization”