Matching is a central problem in pattern recognition and computer vision. A common application is object recognition, detection and tracking. The various matching methods that have been proposed can be distinguished by what type of features is used.
This thesis describes an algorithm for shape matching based on edge feature. As an attempt to achieve accurate edge matching, extended version of Hausdorff distance measure considering edge position and orientation is proposed and orientation-based multiple distance maps are used to compute proposed matching measure efficiently. For the worst case of matching, error compensation technique based on the uncertainty of orientation information is used.
Simulation results show that the proposed algorithm performs a robust matching on images in which edges are dense and noisy. It is shown that application to the ATR system using the proposed matching measure provides good detection result in FLIR still imagery.