<p>Given the advantages of high computational efficiency and low storage cost, hash technology has been extensively researched in the field of cross-modal retrieval. However, the majority of existing methods overlook the richness of latent semantic information and the impact of binary hash codes on semantic extraction. For example, some methods tend to directly decompose latent semantic information from data that aligns with the dimensionality of hash codes and subsequently associate it directly with the hash codes, potentially resulting in a loss of semantic information. To address this issue, we propose an Asymmetric Semantic Preserving Hashing method. Specifically, we use a clustering technique to extract latent semantic information from the original data, incorporating label information into the extraction process to enhance the discriminative power of the latent semantics. Subsequently, we derive low-dimensional compressed semantic information from this latent information via projection, enabling the embedding of latent semantics into hash codes while minimizing the impact of binary representation on semantic extraction. Finally, we embed the compressed semantic information into hash codes using an asymmetric semantic preserving structure, thereby generating highly discriminative hash codes. Extensive experimentation on benchmark datasets has convincingly demonstrated the superiority of our ASPH in terms of retrieval performance.</p>