Segment based image classification

Abstract

Five aspatial and spatial classification methods were compared in this study: standard per‐pixel maximum likelihood classification; Kettig and Landgrebe's ECHO classification; maximum likelihood classification using the segment mean; classification using the segment divergence index; and maximum likelihood classification using the segment probability density function (PDF). The five classification methods were compared using test data from digital aerial imagery with a nominal 1‐m pixel size, and four multispectral bands, acquired over Morgantown, West Virginia, USA. Classification using the segment divergence index produced the lowest accuracy, followed by ECHO, standard maximum likelihood classification and classification with segment mean. The highest accuracy was obtained from classification using the segment PDF.

Acknowledgments

This project was funded by the Ministry of Science and Technology, Republic of Korea. The authors would like to thank two anonymous reviewers for suggestions for improving this paper.