Monogenic bandpass filters for the segmentation of urban zones in satellite sensor imagery

Abstract

This letter introduces a new and effective method for urban zone segmentation in satellite sensor imagery. The proposed method involves filtering a band of a multispectral image with a monogenic bandpass isotropic filter. The filter bandwidth and tuning frequency are determined from the source images by a simple experimental analysis. The final image is obtained by filtering and thresholding the magnitude of the monogenic filter output. Sea zones containing similar features to those found in urban areas are eliminated by a neural network. Experiments with SPOT imagery and comparisons with existing methods are presented as well.