Distributed and Consistent Multi-Image Feature Matching via QuickMatch

TL;DR

This paper extends the QuickMatch multi-image feature matching algorithm to a distributed setting, enabling efficient, consistent matching across many images with minimal communication, demonstrated on low-quality images.

Contribution

It introduces a distributed scheme for QuickMatch that preserves match quality and reduces communication, advancing multi-image feature matching capabilities.

Findings

QuickMatch outperforms standard techniques in accuracy and scale.

Distributed QuickMatch maintains match consistency comparable to centralized version.

Effective on low-quality images, demonstrating robustness.

Abstract

In this work we consider the multi-image object matching problem, extend a centralized solution of the problem to a distributed solution, and present an experimental application of the centralized solution. Multi-image feature matching is a keystone of many applications, including simultaneous localization and mapping, homography, object detection, and structure from motion. We first review the QuickMatch algorithm for multi-image feature matching. We then present a scheme for distributing sets of features across computational units (agents) that largely preserves feature match quality and minimizes communication between agents (avoiding, in particular, the need of flooding all data to all agents). Finally, we show how QuickMatch performs on an object matching test with low quality images. The centralized QuickMatch algorithm is compared to other standard matching algorithms, while the Distributed QuickMatch algorithm is compared to the centralized algorithm in terms of preservation of match consistency. The presented experiment shows that QuickMatch matches features across a large number of images and features in larger numbers and more accurately than standard techniques.