PDF version of this document

NON-RIGID REGISTRATION ASSESSMENT ERRORS

Abstract:

In a recent document that was entitled Error Propagation, detailed explanation was provided, which culminated in the generation of sensitivity plots. These sensitivity plots gave insight into the strengths and validity of non-rigid registration (NRR) assessment methods. We dealt with two such assessment methods, which were rather independent and begged for some means of comparison. Ultimately, a criterion was devised, which could reliably distinguish between the two and show how one method surpassed the other.

A step was missed in that explanation, however, which ought to have fully covered the derivation of some numbers. Primarily, there needed to have been a discussion on how error bars get assigned to values of our assessment methods. Moreover, there ought to be a clearer explanation as to what they practically represent and where they all come from.

This document explains, to some degree, how such values are computed both in the case of model-based assessment, where an appearance model gets evaluated, as well as the case of overlap-based assessment. Since both methods are quite different in nature, calculation of the errors should be treated separately, each in turn.

The closing section briefly describes the steps which follow, but for a fuller, more contextual review, the previous related document ought to be looked up in conjunction.