-
SPS
IEEE Members: $11.00
Non-members: $15.00Length: 0:11:11
Hashing methods have been widely used for similarity search in multimedia due to its low memory requirements and efficient and scalable search. Recently, an approach called Semantic Preserving Hashing (SePH) has been proposed, which uses the semantic probabilities of training data, approximates them with the learnt hash codes and then uses the kernel logistic regression for learning the projection of features to the learnt hash codes. In this paper, we extend this method using a Bayesian framework to learn these projection functions, motivated by the probability distribution that visual features tend to approximate. The proposed Bayesian ridge-based Semantic Preserving Hashing (BiasHash) approach is shown to outperform seven state-of-the-art methods on three benchmark datasets.