Web-Based Planning for Image Processing

Over the past couple years, we have been applying a domain-independent planner, COLLAGE, towards automating Earth science image processing tasks. As reported in previous papers [5, 7], the planner has been successfully used to construct executable image processing data flows. Each component of a data flow corresponds to an image processing operation (e.g., application of a particular image filtering algorithm). Given a particular scientific goal (e.g. "determine vegetation content on the ground for each pixel in this image") and knowledge about the availability and nature of image processing algorithms and how they can be combined to achieve goals, the planner constructs a plan (data flow) which, after being executed, should yield the desired result. The planner can thus be viewed as a tool for automatic program synthesis, or alternatively, as a "component technology" i.e. it constructs programs from given software components. Similar efforts include the Vicar [1] and Amphion [8] systems.