I am trying to classify Landsat 8 MS data for some application. Im using ERDAS IMAGINE 2013 software tool for the same. The tool, however, provides an option for using non-parametric rules (parallepiped and feature space). I am unable to understand how these rules work , and what is their significance in classification. I am observing significant changes in classification accuracy with the use of non-parametric rules. Can someone tell me how to understand these rules. Also, where can I find the mathematical analysis of using these rules.