Discrete matrix factorization cross-modal hashing with multi-similarity consistency

Abstract Recently, matrix factorization-based hashing has gained wide attention because of its strong subspace learning ability and high search efficiency. However, some problems need to be further addressed. First, uniform hash codes can generated by collective factorization, but they often cause serious loss, degrading the quality codes. Second, most them preserve absolute similarity simply in codes, failing capture inherent semantic affinity among training data. To overcome these obstacles, we propose a Discrete Multi-similarity Consistent Matrix Factorization Hashing (DMCMFH). Specifically, an individual is first learned factorization multi-similarity consistency for each modality. Then, subspaces are aligned shared space generate homogenous Finally, iterative-based discrete optimization scheme presented reduce quantization loss. We conduct quantitative experiments on three datasets, MSCOCO, Mirflickr25K NUS-WIDE. Compared with supervised baseline methods, DMCMFH achieves increases $$0.22\%$$ 0.22 % , $$3.00\%$$ 3.00 $$0.79\%$$ 0.79 image-query-text tasks datasets respectively, $$0.21\%$$ 0.21 $$1.62\%$$ 1.62 $$0.50\%$$ 0.50 text-query-image respectively.