Meaningful image objects for object-oriented image analysis

Author:
Publisher website
Google Scholar
Abstract
Image objects obtained by segmentation usually provide a much more reliable representation of real world objects than individual pixels. However, in regions with high quality spatial information, image analysis should focus on objects of interest rather than artificial image objects. A simple method for applying this object-oriented approach consists of converting existing vector geographic information system (GIS) data into raster objects. A problem with this method is that it may produce image objects with saw-toothed edges which barely match measured objects boundaries. In order to address such a problem, a method for adjusting boundaries of image objects is proposed. The new method uses a vector square grid for pixel representation. Vector-based image objects exhibit boundaries which better reproduce the shape and appearance of GIS objects. The proposed approach was applied to extract geometric and biophysical properties of agricultural plots from remotely sensed imagery. Results suggest that vector-based image objects provide much more accurate values than raster-based image objects.