CLASSIFICATION WITH REJECTION PROBLEM FORMULATION AND AN OVERVIEW
Władysław Homenda, Witold Pedrycz · Pattern Recognition · 2018
The chapter offers a detailed formulation of the problem of classification with rejection, delivers a number of illustrative examples, and elaborates on existing studies carried out in this area. It discusses rejecting foreign patterns based on adapted standard recognizers. From the perspective of flow of processing, one can distinguish between three different architectures: rejection prior to classification of native patterns; rejection after classification of native patterns; and rejection simultaneously with classification of native patterns. In order to validate the quality of the proposed rejection mechanisms, several datasets are used, which are to play the role of foreign patterns. The chapter also discusses a particular realization of a binary tree-based classifier for the dataset of handwritten digits. The chapter outlines several concepts on how to employ ensembles of binary classifiers to perform the rejection task.