ACCURACY ASSESSMENT OF IMAGE CLASSIFICATION ALGORITHMS

Abstract

Iso-cluster unsupervised classification was performed using the multivariate toolset of ArcMap 10.1 to identified the spectral clusters or natural statistical groupings present in Kwali Area Council of the Federal Capital Territory Abuja using 2011 Landsat-7 ETM+ and adopting supervised classification that involves ground truthing, the previous knowledge of the study area and creating training site. The maximum likelihood classification (supervised classification), default colour was changed to multiple colour that can easily be interpreted. The new colour assignment was based on information obtained from prior knowledge of the study area. The supervised classified image was further processed to remove all the noises - unwanted or non-relevant information that made it appeared speckled. Using the generalization toolset of ArcMap 10.1 spatial analyst tool, the classified output was filtered to remove the noise; this was done using eight nearest neighbours kernel majority filter. Also, the ragged boundaries of the classified output were smooth as well as clumping the classes together using boundary clean toolset.