Iterative Quantization: A Procrustean Approach to Learning Binary Codes (ITQ)

Problem

Similarity-preserving binary codes for efficient retrieval in large-scale image collections.

ITQ Method

1. Dimensionality Reduction

Such as PCA.

Ommited here.

2. Binary quantization

symbols:
$X\in \mathbb R^{n\times d}$: Data matrix, zero-centered data, i.e., $\sum_{i=1}^n x_i = 0$
$V \in \mathbb R^{n\times c}$: Data matrix after dimensionality reduction.
$R\in \mathbb R^{c\times c}$: Rotation matrix
B∈{ −1,1}n×c