Hashing orthogonal constraint loss for multi-label image retrieval

With the exponential growth of image data on the Internet, large-scale image retrieval has become increasingly important. Hash coding serves as a fundamental technique to achieve efficient retrieval. Traditional deep hashing methods typically optimize the feature distribution by maximizing the inter-class distances. However, when applied to multi-category datasets, they require frequent adjustments of the hash centers, leading to increased training complexity. This paper proposes Hashing Orthogonal Constraint Loss (HOC Loss) to establish a robust metric space that effectively preserves the fine-grained similarity of multi-label images. HOC Loss improves both interclass separation and intraclass aggregation by enforcing orthogonality within the feature space. We conduct experiments on four commonly used datasets to confirm the effectiveness of the method. The results show that our method outperforms existing methods with mean average precision (mAP) improvements of 1.0%, 0.9%, 2.1%, and 2.7%, respectively.