A novel cross-modal hashing algorithm based on multimodal deep learning

With the popularity of multi-modal data on Web, cross media retrieval has become a hot research topic. Existing cross modal hash methods assume that there is a latent space shared by multi-modal features, and embed the heterogeneous data into a joint abstraction space by linear projections. However, these approaches are sensitive to the noise of data, and unable to make use of unlabelled data and multi-modal data with missing values in the real-world applications. To address these challenges, in this paper, we propose a novel Multi-modal Deep Learning based Hashing (MDLH) algorithm. In particular, MDLH adopts deep neural network to encode heterogeneous features into a compact common representation and learn the hash functions based on the common representation. The parameters of the whole model are fine-tuned in supervised training stage. Experiments on two standard datasets show that our method achieves more effective results than other methods in cross modal retrieval.