A pattern modeling techniques for representing, analyzing, and classifying features in data sets resulting from fluid dynamics and combustion diagnostics research, with particular emphasis on turbulent flows will be developed. The primary thrust of the research is directed toward a two component model for flow fields; one being oriented along the major direction of anisotropy, describing the large scale flow; the other being a residual pattern representing the independent small scale flow structures. The separation into global and localized flows facilitates subsequent identification and tracking of features such as eddies. Significant data compression will be achievable due to the decorrelation of the flow patterns. Low level image processing techniques will be developed to meet the challenges posed by the unconventional properties of fluid imagery. A method for low level image segmentation based on a topographic classifier will be investigated. In addition, orthogonal transforms for further compression of the two component flow model will be researched. Data acquired from established laboratories will be used to test our modeling strategy. Spatially and temporally resolved 2D data sets will be used to extend our previous work in feature detection and tracking experiments. We expect the proposed work will meet some of the needs voiced by the combustion and heat transfer community concerning the automation of computer analysis techniques for large fluid flow data sets.
