Supervised Learning of Semantics-Preserving Hashing via Deep Neural Networks for Large-Scale Image Search

This paper presents a supervised deep hashing approach that constructs binary hash codes from labeled data for large-scale image search. We assume that semantic labels are governed by a set of latent attributes in which each attribute can be on or off, and classification relies on these attributes. Based on this assumption, our approach, dubbed supervised semantics-preserving deep hashing (SSDH), constructs hash functions as a latent layer in a deep network in which binary codes are learned by the optimization of an objective function defined over classification error and other desirable properties of hash codes. With this design, SSDH has a nice property that classification and retrieval are unified in a single learning model, and the learned binary codes not only preserve the semantic similarity between images but also are efficient for image search. Moreover, SSDH performs joint learning of image representations, hash codes, and classification in a pointwised manner and thus is naturally scalable to large-scale datasets. SSDH is simple and can be easily realized by a slight modification of an existing deep architecture for classification; yet it is effective and outperforms other unsupervised and supervised hashing approaches on several benchmarks and one large dataset comprising more than 1 million images.