Iterative quantization: A procrustean approach to learning binary codes

摘要: This paper addresses the problem of learning similarity-preserving binary codes for efficient retrieval in large-scale image collections. We propose a simple and alternating minimization scheme finding rotation zero-centered data so as to minimize quantization error mapping this vertices hypercube. method, dubbed iterative (ITQ), has connections multi-class spectral clustering orthogonal Procrustes problem, it can be used both with unsupervised embeddings such PCA supervised canonical correlation analysis (CCA). Our experiments show that resulting coding schemes decisively outperform several other state-of-the-art methods.