SPATIAL POST-PROCESSING OF SPECTRALLY CLASSIFIED VIDEO IMAGES BY A PIECEWISE-LINEAR CLASSIFIER

This study focused on investigating the possibility of combining the p iecewise linear classifier (PLC) with simple spatial post-processing t o improve the classification accuracies of spectrally classified image s. The PLC was employed to classify video images. Spatial post-process ing was then applied to the classified images. This technique was foun d to enhance classification accuracies considerably with only a modera te increase in processing time. Since spatially processed classificati on results were scene-dependent, it was important to choose an appropr iate size of spatial filter for a study area. However, trade-offs betw een classification accuracy and processing speed with increasing size of spatial filter could be determined by trial and error.