Deep Neural Hashing with Adaptive Feature Fusion for Efficient Image Retrieval

Efficient image retrieval is of paramount importance in managing the increasing volume of visual data across diverse domains. Despite advancements in deep learning, extant methodologies frequently encounter challenges in efficient feature integration, adaptability, and contextual comprehension, thereby impacting retrieval performance on large-scale datasets. This research introduces Deep Neural Hashing with Adaptive Feature Fusion (DNH-AFF), a novel framework that enhances deep hashing techniques. DNH-AFF utilizes adaptive feature fusion to dynamically integrate multi-scale features, ensuring a comprehensive semantic representation. A multi-task loss function optimizes hash code generation by enhancing semantic similarity, inter-class separability, and intra-class compactness. Furthermore, a graph-based module captures global contextual relationships to refine retrieval accuracy. Experimental evaluations conducted on CIFAR-10, NUS-WIDE, and MS-COCO datasets demonstrate that DNH-AFF surpasses state-of-the-art methods in mean Average Precision (mAP) and F1-score, affirming its robustness and scalability. These findings establish DNH-AFF as a promising solution for real-world image retrieval applications.