Intelligent Systems Laboratory
Automatic Procedure Construction
We are
designing techniques to automatically construct procedures to achieve low-to-
mid level vision tasks using the operations of mathematical morphology plus a
small set of utility operations. We have started this work by thoroughly
analyzing
twenty different algorithms for real machine vision tasks that have been
reported in
the literature or that some member of our group has been involved with. The
procedures are extremely diverse, including such tasks as inspection of
watch gears,
hot spot detection in IR images, and identification of leads from a SMD component.
We have broken each task into a sequence of primitive operations and then
categorized the subtasks that each operation or subsequence of operations can
achieve. We are now developing the language in which tasks can be
specified and the method by which the sequence of operations and
control structure
are automatically determined from the task specifications. A predicate calculus
knowledge base containing facts and rules about spatial relations and
morphological operations is used by the system.
