A causality-enhanced multiresolution residual learning framework for image retrieval with fast osprey optimization.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Large-scale image collections necessitate advanced content-based image retrieval (CBIR) systems.
• Traditional CBIR methods struggle with high intra-class variation and inter-class ambiguity.
• Existing systems often lack semantic reliability and computational efficiency.

Purpose of the Study:

• To develop a robust and efficient CBIR framework.
• To enhance semantic understanding and interpretability in image retrieval.
• To address limitations of conventional CBIR approaches.

Main Methods:

• Proposed a causality-enhanced multiresolution residual learning framework.
• Utilized a multiscale ResNet 50 backbone for multi-resolution feature extraction.
• Integrated a causal variational autoencoder for latent spatial dependency modeling.
• Employed an enhanced fast osprey optimization algorithm for feature selection.

Main Results:

• Demonstrated consistent improvements over thirteen state-of-the-art methods on CIFAR-10, Oxford Flowers, and Corel 1000 datasets.
• Achieved significant gains in mean average precision and normalized discounted cumulative gain.
• Showcased reduced retrieval time and enhanced robustness against intra-class variation and inter-class ambiguity.

Conclusions:

• The proposed framework offers a robust, efficient, and practical solution for large-scale image retrieval.
• The integration of deep learning, causal modeling, and optimization yields superior performance.
• The approach maintains discriminative consistency across visually similar categories, proving its practical viability.