Ice Floe Identification in Satellite Images Using Mathematical Morphology and Clustering about Principal Curves

• Main Authors: Banfield, Jeffrey D., Raftery, Adrian E.
• Other Authors: WASHINGTON UNIV SEATTLE DEPT OF STATISTICS
• Format: Text
• Language: English
• Published: 1989
• Subjects: Cartography and Aerial Photography; Snow; Ice and Permafrost; Statistics and Probability; *SEA ICE; *SATELLITE PHOTOGRAPHY; *REMOTE DETECTORS; ENVIRONMENTS; EDGES; COMPUTER PROGRAMMING; MORPHOLOGY; ACCURACY; ICE FORMATION; ESTIMATES; IDENTIFICATION; EROSION; CLUSTERING; ICE; IMAGES; ARTIFICIAL SATELLITES; MATHEMATICS; COVERINGS; AUTOMATIC; OFFSHORE STRUCTURES; VOLUME; PROPAGATION; ALGORITHMS; POLAR REGIONS; WUNR661003; Ice; permafrost; Sea ice
• Online Access: http://www.dtic.mil/docs/citations/ADA213854; http://oai.dtic.mil/oai/oai?&verb=getRecord&metadataPrefix=html&identifier=ADA213854

Identification of ice floes and their outlines in satellite images is important for understanding physical processes in the polar regions, for transportation in ice-covered seas and for the design of offshore structures intended to survive in the presence of ice. At present this is done manually, a long and tedious process which precluded full use of the great volume of relevant images now available. We describe an automatic and accurate method for identifying ice floes and their outlines. Floe outlines are modeled as closed principal curves (Hastie and Stuetzle, 1989), a flexible class of smooth non- parametric curves. We propose a robust method of estimating closed principal curves which reduces both bias and variance. Initial estimates of floe outlines come from the erosion-propagation (EP) algorithm, which combines erosion from mathematical morphology with local propagation of information about floe edges. The edge pixel from the EP algorithm are grouped into floe outlines using a new clustering algorithm. This extends existing clustering methods by allowing groups to be centered about principal curves rather than points or lines. This may open the way to efficient feature extraction using cluster analysis in images more generally. The method is implemented in an object-oriented programming environment for which it is well suited, and is quite computationally efficient. Keywords: Erosion; Feature extraction; LANDSAT; Non- parametric curves.