Autonomous Extraction of an Eddy-Like Structure from Infrared Images of the Ocean

The amount of imagery generated by all the various satellite-borne thermal infrared sensors is great and constantly growing. To analyze this imagery manually is tedious and time consuming. The present paper is concerned with automating the extraction of one important, though often weak, type of oceanic feature viz. isolated blob structure. These "blobs" usually correspond to distinct oceanographic features such as eddies or regions of upwelling. The technique proposed to achieve this aim is based on finding the structural relationship between connected regions of constant grey level (i.e., constant temperature). These elementary connected regions are known as atoms. The relationship between atoms is described by a region adjacency graph with the position on the graph being determined by the depth of enclosure of the particular atom. An eddy-like structure is extracted by searching the graph for isolated atoms of high temperature that are enclosed by atoms of lower temperatures (for warm core eddies). Techniques of preprocessing to reduce the search problem to managable proportions and methods to deal with cloud and land "contamination" are also described. The examples given are based on thermal imagery from the Heat Capacity Mapping Mission (HCMM).