Binary Image Operator Design Based on Stacked Generalization

Abstract

Stacked generalization refers to any learning schema that consists of multiple levels of training. Level zero classifiers are those that depend solely on input data while classifiers at other levels may use the output of lower levels as the input. Stacked generalization can be used to address the difficulties related to the design of image operators defined on large windows. This paper describes a simple stacked generalization schema for the design of binary image operators and presents several application examples that show its effectiveness as a training schema.