Improving Deep Hashing Performance for Image Retrieval Using Multi-Dimensional Attention

Deep learning-based hashing techniques have shown remarkable performance in large-scale image retrieval by efficiently compressing high-dimensional feature data into compact binary codes, which greatly enhance retrieval speed and efficiency. However, as the depth of convolutional neural networks (CNNs) increases, important image information can be weakened or lost, leading to a decrease in retrieval accuracy and limiting practical applications. To address this issue, this paper proposes a novel deep hashing model based on a multi-dimensional attention mechanism—Deep Hashing with Multi-Dimensional Attention (DH-MDA). The proposed model integrates a multi-dimensional attention mechanism to selectively focus on and enhance important feature regions in images, dynamically adjusting their weights based on their relevance across different aspects. This approach effectively mitigates information loss caused by deep network structures and strengthens the representational capacity of the hash codes. To evaluate the performance of the proposed model, extensive experiments were conducted on the CIFAR-10 dataset. The results show that the DH-MDA model significantly improves the mean Average Precision (mAP) of retrieval, achieving a score of 0.788, which underscores the importance of multi-dimensional attention in enhancing deep hashing performance. Additionally, the model demonstrates strong generalization ability, offering valuable insights and technical references for future research in high-dimensional data retrieval.