Fast, Robust Pattern Matching

  • Brandner, Markus, (Co-Investigator (CoI))
  • Pinz, Axel (Principal Investigator (PI))

Project: Research project

Description

A vast number of industrial applications of digital image analysis requires the location of previously trained patterns in large search images in "real time". This problem of "pattern matching" is addressed in a flexible and robust manner. The main goal of the project is to develop a new software library, which can be adapted to many industrial application scenarios. The specifications of the project are as follows:

* Pattern localization accuracy: greater 1/10 pixel

* Invariant to rotations (0-360 degrees)

* Invariant to scaling (0.5 - 2)

* The system should tolerate up to 50% of missing parts

* Invariant to changes in gain and offset of the illumination

* "real-time" is defined as locating a pattern (template greater 128 x 128 pixel) in less than 300 msec.

* Invariant to quasi-affine distortions.
StatusFinished
Effective start/end date1/10/9931/12/03