Probability Ordinal-Preserving Semantic Hashing

Existing semantic hashing methods primarily concentrate on preserving piecewise class information or pairwise correlations in learned binary codes, often neglecting the mutual triplet-level ordinal structure crucial for similarity preservation. This chapter introduces a groundbreaking approach–the Probability Ordinal-preserving Semantic Hashing (POSH)Probability Ordinal-preserving Semantic Hashing (POSH) framework–pioneering ordinal-preserving hashing under a non-parametric Bayesian theory. The framework derives the entire learning process for ordinal similarity-preserving hashing based on maximum posterior estimation. Probabilistic ordinal similarity preservation, probabilistic quantization function, and probabilistic semantic-preserving function are seamlessly integrated into a unified learning framework. Notably, the proposed triplet-ordering correlation preservation scheme enhances the interpretability of learned hash codes using an anchor-induced asymmetric graph learning model. Additionally, a sparsity-guided selective quantization function minimizes space transformation losses, while a regressive semantic function enhances the flexibility of formulated semantics in hash code learning. The joint learning objective concurrently preserves the ordinal locality of original data and explores semantics for discriminative hash code generation. An efficient alternating optimization algorithm, backed by a strictly proof convergence guarantee, is developed to solve the resulting objective problem. Extensive experiments across several large-scale datasets validate the superior performance of the proposed POSH framework compared to state-of-the-art hashing-based retrieval methods.