Supervised locality preserving hashing

Abstract

Hashing methods are becoming increasingly popular because they can achieve fast retrieval of large-scale data by representing the images with binary codes. However, the traditional hashing methods tend to obtain the binary codes by relaxing the discrete problems which greatly increase the information loss. In this paper, we propose a novel hash learning method, called Supervised Locality Preserving Hashing (SLPH) for image retrieval. Different from the traditional two-steps methods which learn low-dimensional features and binary codes of the data separately, we directly obtain the binary codes and thus reduce the information loss. Besides, we add graph-regularized learning on the designed model to avoid over-fitting and improve the performance. Experiments on two benchmark databases show that the proposed SLPH performs better than some state-of-the-art methods.