Object Oriented Digital Image Classification

要旨

Land cover/use mapping at national level using visual classification has been widely accepted in the developing world due to its high classification accuracy (>85%). However this visual approach is time consuming and most countries can ill afford to update their database regularly. Digital supervised classification with lower accuracy and a characteristics salt and pepper appearance has recently been enhanced using the object oriented classification approach based on fuzzy logic algorithms. This study has compared the classification quality of a conventional supervised pixel based classification with that of the object segmented classification using Landsat ETM data sets. The results indicated that the object oriented classification was superior to supervised classification in several aspects; most importantly it gave higher accuracy and did not have the characteristic salt and pepper appearance.