Abstract

Dehvari A.* and Heck R. J.

An airborne infrared image was used to produce a map of land cover types in the Eastern shore of Lake Huron, Ontario province of Canada. Maximum likelihood pixel-based and nearest neighbor object-based methods were used in this approach. Land cover classes that obtained traditional pixel-based classification approaches showed a saltand- pepper effect having the lowest producer accuracy (59.5%). Overall classification results increased up to 80% in object- based approach but still failed to distinguish buildings and creeks. Contours and DEM thematic layers enhanced classification results to a higher level (94%) and increased the producer accuracy for buildings and creek by creating reasonable objects in segmentation process in the object-based approach.