Color Image Set Recognition Based on Quaternionic Grassmannians

TL;DR

This paper introduces a novel color image set recognition method leveraging quaternionic Grassmannians to effectively capture color information, demonstrating promising results on benchmark datasets.

Contribution

The paper presents a new quaternionic Grassmannian-based framework for color image set recognition, including a direct distance formula and a classification approach.

Findings

Achieves good recognition accuracy on ETH-80 and Highway Traffic datasets.

Provides a direct formula for shortest distance on quaternionic Grassmannian.

Discusses limitations and potential improvements in stability.

Abstract

We propose a new method for recognizing color image sets using quaternionic Grassmannians, which use the power of quaternions to capture color information and represent each color image set as a point on the quaternionic Grassmannian. We provide a direct formula to calculate the shortest distance between two points on the quaternionic Grassmannian, and use this distance to build a new classification framework. Experiments on the ETH-80 benchmark dataset and and the Highway Traffic video dataset show that our method achieves good recognition results. We also discuss some limitations in stability and suggest ways the method can be improved in the future.