When is a confidence measure good enough?

Abstract

Confidence estimation has recently become a hot topic in image processing and computer vision. Yet, several definitions exist of the term "confidence" which are sometimes used interchangeably. This is a position paper, in which we aim to give an overview on existing definitions, thereby clarifying the meaning of the used terms to facilitate further research in this field. Based on these clarifications, we develop a theory to compare confidence measures with respect to their quality. © 2013 Springer-Verlag.