Fractional Multi-view Hashing with Semantic Correlation Maximization

Hashing has been extensively concerned in multimedia research due to its low storage cost and high retrieval efficiency. The purpose of multi-view hashing is to project heterogeneous data from different views in high dimensional space into compact and discrete binary codes in Hamming space and meanwhile maintains the similarity of the original data. In this paper, we propose a Fractional Multi-view Hashing with Semantic Correlation Maximization (FMH-SCM), where the learning objective is to seek the maximum semantic correlation from various views. The proposed method uses labels to learn multi-view hash codes and fractional-order embedding to reduce the negative effect of noise. Moreover, a sequential optimization is used to solve the hash functions for improving the performance of this method. FMH-SCM is compared with related algorithms on two image retrieval datasets. Extensive experiments verify the effectiveness of the proposed method.