Correlated Wishart Matrices Classification via an Expectation-Maximization Composite Likelihood-Based Algorithm

TL;DR

This paper introduces an Expectation-Maximization composite likelihood algorithm for classifying correlated Wishart matrices, improving image set classification by accounting for correlations in positive-definite matrix data.

Contribution

It presents a novel EM-based method for modeling correlated Wishart matrices, addressing a key challenge in classifying RCDs in computer vision.

Findings

The proposed algorithm outperforms existing methods on synthetic data.

It achieves better classification accuracy on real face data.

The method effectively models correlation among RCDs.

Abstract

Positive-definite matrix-variate data is becoming popular in computer vision. The computer vision data descriptors in the form of Region Covariance Descriptors (RCD) are positive definite matrices, which extract the key features of the images. The RCDs are extensively used in image set classification. Some classification methods treating RCDs as Wishart distributed random matrices are being proposed. However, the majority of the current methods preclude the potential correlation among the RCDs caused by the so-called non-voxel information (e.g., subjects' ages and nose widths, etc). Modeling correlated Wishart matrices is difficult since the joint density function of correlated Wishart matrices is difficult to be obtained. In this paper, we propose an Expectation-Maximization composite likelihood-based algorithm of Wishart matrices to tackle this issue. Given the numerical studies based on the synthetic data and the real data (Chicago face data-set), our proposed algorithm performs better than the alternative methods which do not consider the correlation caused by the so-called non-voxel information. All these above demonstrate our algorithm's compelling potential in image set classification in the coming future.