Region adjacency analysis of remotely-sensed imagery
D.G. Nichol · International Journal of Remote Sensing · 1990
The analysis of image data by region adjacency methods is a long-established, though not currently widely-used, method of region analysis. This lack of use is probably because the technique has usually been applied only to images which have been subjected to many stages of processing and which, as a result, contain only a small number of regions whose analysis could easily be achieved manually. The present paper proposes that there are significant benefits from applying region adjacency analysis to comparatively raw (‘low-level’) imagery, such as is produced by many remote sensing systems, rather than to highly processed images. In fact many high-level processing requirements can be better performed by region analysis of the low-level data. Examples include blob extraction, specific region neighbour searching and region merging. The latter operation is extremely important in reducing complex image-derived data to a less complex form suitable for entry into a geographical information system (GIS). A potential difficulty with achieving this analysis with low-level images is that these, in general, have many more regions and a complete analysis, using existing techniques, would be extremely time consuming. In order to overcome this an efficient set of algorithms have been developed which produce the region adjacency graph of a complicated image; this is then used to carry out the required region analysis. The examples of region analysis given include region merging of a noisy space-borne imaging radar picture from the Shuttle Image Radar-B (SIR-B) experiment and contextual region searching, as required for land-use inventory for example, from a spectrally classified Landsat image.