Assessment of accuracy for soft classification: SCAASFER as a tool

Abstract

Accuracy assessment is an important issue for validation of image classification. There are several tools that provide accuracy analysis using confusion matrix, Kappa coefficients and others incorporated in ERDAS_IMAGINE, ArcGIS, ENVI, Geomatica etc. There are some of tools available for soft or sub-pixel classification which gives fraction images as output like, FUZCEN, SMIC. Each fraction image is associated to single class from classification schema and represents membership to that class in gray scale for all pixels of input multispectral image. This paper introduces SCAASFER: Soft Classification Accuracy Assessment using SCM, FERM, Entropy and RMSE as a tool for accuracy assessment for soft classified data. This tool automated the accuracy assessment process with several popular accuracy measures available for soft classification, like FERM (Fuzzy Error Matrix), SCM (Sub-pixel Confusion uncertainty Matrix), MIN-PROD Error matrix, RMSE (Root Mean Square Error) and Entropy. SCAASFER provide an interactive way to acquire samples data for accuracy from fraction images with different options, like Random points, stratified random points, manual selection of points and area from fraction image or take compete image census for accuracy analysis. This tool also provides facility to view and save accuracy assessment results as well as sample points on hard disk. This tool is developed in JAVA programming language. JAVA is a platform independent language and allows running this tool on different operating Systems and machines.