Some Fuzzy Tools for Evaluation of Computer Vision Algorithms

Abstract

In this article, some issues related to the performance evaluation of computer vision algorithms within the version of direct empirical supervised evaluation method developed at SRISA RAS are considered. This approach partly relies on the elements defined by using the fuzzy set theory, in particular, fuzzy similarity measures and fuzzy reference ground truth images. Some known measures of segmentation quality are considered and their extensions, representing the fuzzy similarity measures, are offered. As an example, the author considers an application of fuzzy ground truth images and fuzzy similarity measures, including some newly introduced ones, to the evaluation of face recognition algorithms.