Automatic Recognition of Changes in Urban Development from Aerial Photographs

Abstract

A computer is programmed to detect changes in two aerial photographs of the same region taken at different times. A nonlinear registration technique is used to partition the photographs into cell pairs. Three feature measurements called the two-dimensional correlation coefficient, the average entropy change per picture element, and the high-intensity probability change per picture element are calculated for each cell pair. The pattern recognition system is trained on a set of learning samples and then tested on an independent set of test samples. Experimental error probability curves are presented as a measure of the system effectiveness.