Using the Correspondence Framework to Select Surface Matching Algorithms
Authors: B.M. Planitz, A.J. Maeder, J.A. Williams
Date: February 2005
Abstract:
A correspondence framework has recently been proposed to unify a wide variety of surface matching algorithms, and provide a consistent structure for establishing new ones. When an algorithm is implemented using the framework, it is divided into five stages. A module is created for each stage of the framework, and that module is placed in a library (for that stage of the framework). Algorithms are created by connecting five appropriate modules from the library. It is envisaged that in the future, algorithms will be created by automatically connecting five suitable modules for their specific surface matching tasks. This paper takes a step towards this goal, by presenting a metric for assessing the outcomes of the final stage of the framework. The metric provides a quantitative value that determines the suitability of an algorithm for a specific task. Six algorithms are presented and their suitability over a range of surfaces is tested. Results show that the outcome of each experiment reflects the expected outcome. Thus, the metric is an appropriate tool for algorithm selection. Future directions at the end of the paper discuss the concept of using metrics at the other stages of the framework, so that the automatic algorithms selection process can be realised.
Keywords: Correspondence, surface matching, registration, object recognition
Publisher: © 2005 APRS. Proceedings of the Australian Pattern Recognition Society (APRS) Workshop on Digital Image Computing (WDIC 2005), Brisbane, Australia, 12 February 2005, CD-ROM pp 85-90
Download the paper (PDF: 814KB)