Binary Representation via Jointly Personalized Sparse Hashing

Unsupervised hashing has attracted much attention for binary representation learning due to the requirement of economical storage and efficiency codes. It aims encode high-dimensional features in Hamming space with similarity preservation between instances. However, most existing methods learn hash functions manifold-based approaches. Those capture local geometric structures (i.e., pairwise relationships) data, lack satisfactory performance dealing real-world scenarios that produce similar (e.g., color shape) different semantic information. To address this challenge, work, we propose an effective unsupervised method, namely, Jointly Personalized Sparse Hashing (JPSH), learning. be specific, first, a novel personalized module, i.e., (PSH). Different subspaces are constructed reflect category-specific attributes clusters, adaptively mapping instances within same cluster space. In addition, deploy sparse constraints select important features. We also collect strengths other clusters build PSH module avoiding over-fitting. Then, simultaneously preserve similarities our proposed JPSH, incorporate into seamless formulation. As such, JPSH not only distinguishes from but preserves neighborhood cluster. Finally, alternating optimization algorithm is adopted iteratively analytical solutions model. apply search task. Extensive experiments on four benchmark datasets verify outperforms several state-of-the-art algorithms.