Spatial pattern recognition by means of representativeness measures

Spectral pattern recognition methods have been widely used in remote sensing, especially in pixel-by-pixel approaches. Spatial properties of satellite imagery have been exploited to a much lesser extent. Common features to be used for the latter include texture, shape, directionality etc. A spatial property which has not been used for image segmentation purposes is the spatial representativeness of each pixel. The approach described herein combines image segmentation and per-field classification. It is intended to be used for the classification of land-cover or vegetation respectively. The initial segmentation by means of representativeness is the main feature of this approach