Measures based on fuzzy correlation coefficient for accuracy assessment of fuzzy classification

Abstract

In this study, measures based on fuzzy correlation co-efficient have been used to assess the accuracy of fuzzy classification. The measures make use of category membership value of a class in a pixel with respect to its sub-pixel reference data. Different measures have been defined to find accuracy assessment of four different fuzzy classification outputs i.e., accuracy assessment of (i) a single class in a single pixel, (ii) a single class in all the samples, (iii) sub-pixel classification (all the classes in the same pixel) and (iv) an image (considering all the classes in all pixels). These along with Pearson's correlation coefficient have been used to test fuzzy classification of IRS LISS III PAN data by two different classifiers. It has been found that all the measures, including Pearson's correlation coefficient, assessed the same classifier better than the other. Thus, it can be concluded that measures based on fuzzy correlation co-efficient can be used to assess the accuracy of fuzzy classification. However, a comparison of the capability of thefuzzy correlation measures with that of existing measures may be further be considered as a future scope of study. © Geoinformatics International.