Using AI Planning Techniques to Automatically Generate Image Processing Procedures: A Preliminary Report

Steve Chien

This paper describes work on the Multimission VICAR Planner (MVP) system to automatically construct executable image processing procedures for custom image processing requests for the JPL Multimission Image Processing Lab (MIPL). This paper focuses on two issues. First, large search spaces caused by complex plans required the use of hand encoded control information. In order to address this in a manner similar to that used by human experts, MVP uses a decomposition-based planner to implement hierarchical/skeletal planning at the higher level and then uses a classical operator based planner to solve subproblems in contexts defined by the high-level decomposition. Second, the image processing domain is characterized by large amounts of search to find the correct program options for images (e.g. operator effects), rather than search among different programs (e.g. planning operators) and many of these program options are incompatible (i.e. certain combinations cannot be used). MVP represents these interactions by using codesignation constraints to specify program options for operators and allowing these constraints to occur in operator preconditions allowing MVP to search the program option space efficiently while handling negative interactions between program options.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.