Automatic interpretation of images from Earth Resources Technology Satellite-1 (ERTS-1, now called LANDSAT) can be used in a variety of applications with considerable accuracy. Most systems, however, classify strictly on a point by point basis, making no use of any spatial knowledge. Standard photo-interpretation techniques are combined with some techniques from artificial intelligence to produce an increase in accuracy over a point-by-point classification method. Traditional classification methods are used to obtain an initial segmentation of the image. Then, a controlled region merging process allows the regions with unambiguous interpretations to influence the interpretation of neighbouring ambiguous regions, thereby introducing considerable context sensitivity into the interpretation process. Results are given of an experiment to interpret areas of different forest cover. / Science, Faculty of / Computer Science, Department of / Graduate