Enhanced visual content retrieval via hierarchical deep fusion of multi-level CNN features

초록

Convolutional Neural Networks (CNNs) have emerged as a leading method for extracting deep image features, essential for accurate image retrieval. This paper introduces a versatile hierarchical deep fusion architecture that enhances CNNs' discriminative power by integrating multi-level information fusion-algorithmic, signature-based and semantic. The proposed method incorporates cost-effective, context-sensitive modules to retrieve semantically similar images from cluttered or overlapping environments. Experimental results on eleven benchmark datasets, including Caltech-101, CIFAR-10, and Corel-1000, demonstrate the effectiveness of the presented method, achieving advanced performance with improved Mean Average Precision (mAP), recall, and retrieval efficiency.

키워드