PDF version of this entire document

Normalisation Experiments: Planning, Deployment, and Analysis

Roy Schestowitz

Abstract:

This document outlines an augmented experimental plan. It contains a succinct proposition for experiments to be run, which in turn help us understand how to normalise assessors of non-rigid registration (NRR) that use appearance models. The core of these experiments involves two synthetic (randomly-drawn) distributions which are deliberately varied for the purpose of subsequent analysis. Additionally, different factor in the evaluation, such as the power to which differences get raised, are being investigated. The outcome of the experiments is of special interest in its own right, yet it also helps provide a more objective measure for real-world applications and makes the evaluation framework we have more valuable.