In this paper we discuss uses of image segmentation, feature extraction and
Bayesian networks for identifying buildings in digital orthophotos and the
utilisation of the results for the automated computation of building stati
stics. Our work differs from previous attempts in a number of ways. Firstly
, image segmentation is accomplished using an adaptive multi-scale method w
hich brings together region and edge information to segment the image into
regions. Secondly, automated building feature extraction (e.g. corners) is
optimised to fit with expert human annotation performance. The third aspect
of this work is the exploration of Bayesian networks as a method for fusin
g available information (ranging from corner information to solar angles as
indicators of shadow location) to classify segmented regions as correspond
ing to buildings or not. Such processes then permit the automatic compilati
on of building statistics.