Binary Auto-Encoders Hashing with Manifold Similarity-preserving for Image Retrieval

Abstract

By minimizing reconstruction loss, binary auto-encoders algorithm makes the hash codes keep the important information of the original input. However, the algorithm only considers the reconstruction loss and doesn't consider the local geometric structure information between the original data, which is bad for learning high quality hash functions. Therefore, in this paper, we propose a new image retrieval algorithm based on binary auto-encoders hashing with manifold similarity-preserving (MSP-BAH). First, the supervised Laplacian eigenmaps algorithm for the generation of the referenced hash codes is used to generate referenced hash codes, which keep the local geometric structure information of the original input data. Then by using the hash codes as a reference, the manifold similarity-preserving loss is constructed between the hash codes generated by the encoder and the referenced hash codes to guide the learning process of the model, so that the MSP-BAH model can provide strong characterization ability while keeping the local geometric structure information unchanged as much as possible. We perform some experiments on three benchmark datasets, i.e., CIFAR10, MNIST, and NUS-WIDE. The results show that the MSP-BAH method has better performance comparing with several existing image retrieval methods.