Hypergraph-Based Discrete Hashing Learning for Cross-Modal Retrieval

Anchor Graph Structure Fusion Hashing for Cross-Modal Similarity Search

“…By whether the label information is utilized, cross-modal hashing (CMH) methods can be categorized into two subclasses: unsupervised [15][16][17][18][19][20] and supervised [21][22][23][24][25] methods. The details are shown in Section II.…”

“…Firstly, most existing CMH methods take graphs, which are always predefined separately in each modality, as input to model the distribution of data. These methods omit to consider the correlation of graph structure among multiple modalities, and cross-modal retrieval results highly rely on the quality of predefined affinity graphs [15,16,18]. Secondly, most existing CMH methods deal with the preservation of intra-and inter-modal affinity separately to learn the binary code, omitting to consider the fusion affinity among multimodalities data containing complementary information [16][17][18].…”

“…These methods omit to consider the correlation of graph structure among multiple modalities, and cross-modal retrieval results highly rely on the quality of predefined affinity graphs [15,16,18]. Secondly, most existing CMH methods deal with the preservation of intra-and inter-modal affinity separately to learn the binary code, omitting to consider the fusion affinity among multimodalities data containing complementary information [16][17][18]. Thirdly, most existing CMH methods relax the discrete constraints to solve the optimization objective, which could significantly degrade the retrieval performance [16,21,22].…”

“…Unsupervised CMH methods [15][16][17][18][19][20] mainly map the heterogeneous multi-modal data instances into the common compact binary codes by preserving the intra-and inter-modal relevance of training data and learn the hash functions during this process. Fusion Similarity Hashing (FSH) [15] learns binary hash codes by preserving the fusion similarity of multiple modalities in an undirected asymmetric graph.…”

“…Joint Coupled-Hashing Representation (JCHR) [17] learns the unified hash codes via embedding heterogeneous data into their corresponding binary spaces. Hypergraph-based Discrete Hashing (BGDH) [18] obtains the binary codes with learning hypergraph and binary codes simultaneously. Robust and Flexible Discrete Hashing (RFDH) [19] directly produces the hash codes via discrete matrix decomposition.…”