UBC Theses and Dissertations
Using image hierarchies to interpret LANDSAT data Catanzariti, Ezio
Most automatic LANDSAT image interpretation systems have used traditional Pattern Recognition techniques. Usually each pixel is classified into one of a number of categories by examining its spectral signature, without regard to its spatial context. A survey of such techniques and of computational vision techniques from Artificial Intelligence leads to the design of a new system that allows the spatial structure of the image to control the interpretation. This classifier uses a pyramidal, hierarchical structure of images. A number of experiments with the implementation on LANDSAT images of forest cover show that one can achieve improvements over a conventional classifier in both accuracy (number of pixels correctly interpreted) and readability (number of regions in the interpreted image) without sacrificing efficiency.
Item Citations and Data