Object-based classification using Quickbird imagery for delineating forest vegetation polygons in a Mediterranean test site

A multi-scale, object-based analysis of a Quickbird satellite image has been carried out to delineate forest vegetation polygons in a natural forest in Northern Greece. Following a multi-resolution segmentation, a classification tree was developed and compared using a nearest neighbour classifier for the assignment of image segments to classes. Additionally, texture images derived from local indicators of spatial association were calculated and used to improve the classification. st results were obtained when texture images were considered in the classification sequence, however, the accuracy of the final map did not exceed 80%. The classification tree yielded better results than the nearest neighbour algorithm. Overall, the object-based classification approach presented both advantages and limitations, which have to be considered prior to its operational use in mapping Mediterranean forest ecosystems.