Knowledge-based adaptive identification of 2D image structures
Josef Pauli · Close-Range Photogrammetry Meets Machine Vision · 1990
This paper presents a procedure for automatically fine-tuning the control parameters of a noise-tolerant matching algorithm. The construction of correspondence relations between 2D models and image representations will be executed by computing morphisms between relational structures. Tolerance parameters for model attributes can be modified in order to apply the matching algorithm to real world images. Evaluation functions are incorporated for measuring the quality of the correspon- dence relations. An adaptation procedure automatically modifies the tolerance parameters to reach acceptable corresponden- ce relations. For initializing the adaptation procedure the user only has to specify initial values of a minimum set of tolerance parameters. Geometrical dependencies between attributes of model components are employed for determining initial values for the rest of the tolerance parameters. Additionally, the user has to define several criteria for accepting correspondence relations. These acceptance criteria will be specified by fixing thresholds for the evaluated correspondence relations. Several modification strategies are included for reacting appropriately on the outcome of the evaluation functions. The system will be demonstrated in the practical application area of inspecting classes of Integrated Circuits to detect manufacturing errors.