Supervised discrete hashing for hamming space retrieval

Recently, Hamming space retrieval, which realizes the foremost effective constant-time search, is incredibly prevalent in large-scale approximate neighbor retrieval since its high computational potency. However, notwithstanding the exciting progress, the accuracy of extant hashing strategies for Hamming space retrieval continues to be faraway from satisfactory. To handle this issue, this research proposes a novel discrete hashing method named Supervised Hamming Hashing (SHH). Specifically, supervision is carefully tailored to reduce the semantic classification loss, which incorporates label regression strategies and asymmetric similarity preserving scheme. Further, a unique hash code fusion strategy and a customized discrete optimization algorithm are designed for optimizing hash codes, thus heightening the potency and precision of the hash codes. A large number of experiments carried out on three benchmarks have corroborated the advantageous performance of SHH in Hamming space retrieval.(c) 2022 Elsevier B.V. All rights reserved.