Image Retrieval via Canonical Correlation Analysis

Abstract

Canonical Correlation Analysis (CCA) is a powerful multivariate statistical method. It can be used to find, for a given dimension, a projection pair that maximally captures the correlation between two target random vectors. This work introduces a CCA-based approach for image retrieval. It capitalizes on feature maps extracted from a pre-trained Convolutional Neural Network (CNN) and leverages basis vectors identified via CCA, in conjunction with an element-wise selection method based on the Chernoff information, to generate compact transformed image features; the level of similarity between two images is determined by a hypothesis test regarding the joint distribution of transformed feature pair. The proposed approach is benchmarked against two popular statistical analysis methods, Linear Discriminant Analysis (LDA) and Principal Component Analysis with whitening (PCAw). The CCA approach is shown to achieve competitive retrieval performances on popular datasets such as Oxford5k and Paris6k.

Publication
16th Canadian Workshop on Information Theory (CWIT)