Cluster-wise unsupervised hashing for cross-modal similarity search

Abstract Cross-modal hashing similarity retrieval plays dual roles across various applications including search engines and autopilot systems. More generally, these methods also known to reduce the computation memory storage in a training scheme. The key limitation of current are that: (i) they relax discrete constrains solve optimization problem which may defeat model purpose, (ii) projecting heterogenous data into latent space encourage loss diverse representations such data, (iii) transforming real-valued point binary codes always resulting information producing suboptimal continuous space. In this paper, we propose novel framework project original points from different modalities its own low-dimensional finds cluster centroid space, using Cluster-wise Unsupervised Hashing (CUH). particular, proposed clustering scheme aims jointly learns compact hash corresponding linear functions. A is developed learn unified under guidance cluster-wise code-prototypes. Extensive experiments over multiple datasets demonstrate effectiveness our comparison with state-of-the-art unsupervised cross-modal tasks.