A line-preserving post-processing technique for maximum likelihood classification of SAR images

Abstract

Speckle in Synthetic Aperture Radar (SAR) images causes many pixels to be misclassified when a maximum likelihood classifier (MLC) is applied. We present a post-processing procedure for removing these misclassified pixels. The procedure, requiring only a few comparisons for each pixel, first detects misclassified pixels, then reclassifies them to the classes of their neighbouring pixels. Experimental results show that this procedure efficiently reduces misclassified pixels while preserving lines, boundaries and corners in thematic maps.