Weighted multi-deep ranking supervised hashing for efficient image retrieval

Deep hashing has proven to be efficient and effective for large-scale image retrieval due to the strong representation capability of deep networks. Existing deep hashing methods only utilize a single deep hash table. In order to achieve both higher retrieval recall and precision, longer hash codes can be used but at the expense of higher space usage. To address this issue, a novel deep hashing method is proposed in this paper, weighted multi-deep ranking supervised hashing (WMDRH), which employs multiple weighted deep hash tables to improve precision/recall without increasing space usage. The hash table is constructed as an additional layer in a deep network. Hash codes are generated by minimizing the loss function that contains two terms: (1) the ranking pairwise loss and (2) the classification loss. The ranking pairwise loss ensures to generate discriminative hash codes by penalizing more for the (dis)similar image pairs with (small)large Hamming distances. The classification loss guarantees the hash codes to be effective for category prediction. Different hash bits in each individual hash table are treated differently by assigning corresponding weights based on information preservation and bit diversity. Moreover, multiple hash tables are integrated by assigning the appropriate weight to each table according to its mean average precision (MAP) score for image retrieval. Experiments on three widely-used image databases show the proposed method outperforms state-of-the-art hashing methods.