Stable Feature Extraction in Aerial Reconnaissance Images

摘要

The problem of extracting stable features in aerial reconnaissance images is addressed in this report. Fractal models of image features are proposed an concept of fractal error is introduced. Fractal error is a metric that quantifies the 'closeness of fit' of the image feature to a fractal model. Using this metric one can discriminate between natural and man-made features in an image. Fractal error also presents a useful approach to extracting edges in images. Methods are given for computing and approximating fractal error using networks and genetic algorithms.