Hash Learning with Convolutional Neural Networks for Semantic Based Image Retrieval

Jinma Guo, Shifeng Zhang, Jianmin Li

Abstract: Hashing is an effective method of approximate nearest neighbor search (ANN) for the massive web images. In this paper, we propose a method that combines convolutional neural networks (CNN) with hash learning, where the features learned by the former are beneficial to the latter. By introducing a new loss layer and a new hash layer, the proposed method can learn the hash functions that preserve the semantic information and at the same time satisfy the desirable independent properties of hashing. Experiments show that our method outperforms the state-of-the-art methods by a large margin on image retrieval. And the comparisons with baseline models show the effectiveness of our proposed layers.