Discriminant Cross modal Hash Retrieval Algorithm with Multilevel Semantics

In view of the fact that most of the cross modal hash methods use binary matrices to represent the degree of correlation, they cannot capture the deeper semantic information between multi label data, and they ignore the problems of maintaining the discriminability of semantic structure and data characteristics, so a discriminant cross modal hash retrieval algorithm based on multi-level semantics, ML-SDH, is proposed. The proposed algorithm uses multi-level semantic similarity matrix to find the deep correlation information in the cross modal data, and uses the equal guidance cross modal hash to represent the correlation in the semantic structure and discrimination classification, which realizes the purpose of encoding the multi tag data containing high-level semantic information. Moreover, the structure that preserves multi-level semantics can ensure that the final learned hash code is discriminative while maintaining semantic similarity. On the NUSWIDE dataset, when the hash code length is 32bit, the average accuracy (mAP) of the proposed algorithm in the two retrieval tasks is 19.48, 14.50, 1.95 percentage points higher than that of the deep cross-modal hashing (DCMH), paired associative hash (PRDH), and equality guided discriminant hash (EGDH) algorithms, respectively, and 16.32, 11.82, 2.08 percentage points higher than that of the proposed algorithm.