Abstract
In recent years hashing methods have attracted considerable attention. The main procedure of manifold hashing methods is to convert the high dimension features into binary codes and then use manifold distance to approximate the original Euclidean distance. However, there still is a problem that how to directly preserve the manifold structure by hashing. Besides, because of information loss, the effect of using existing hashing methods to retrieve images is not ideal when the length of binary code is very short. In order to solve these problems, we proposed a strong constrained manifold-embedded hash (SCDMH), which can not only explore the nonlinear manifold structure of data, but also make full use of the semantic information in the data. Additionally, the objective function we proposed reconstructs the original features of the datasets and the hash representation mutually to preserve the semantic information. Our experimental results on three large benchmark datasets demonstrate that SCDMH outperforms other state-of-the-art baselines.
Export citation and abstract BibTeX RIS
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.