Object recognition using a dynamic relaxation method

Abstract

In this paper, we show a novel approach for solving the classical feature matching problem between a pair of predicted and extracted images. Point features are considered, and the feature matching is performed using a dynamic relaxation method. The proposed method examines the evidence in the local neighborhoods of a possible matching predicted and extracted feature pair and determines the validity of the current match. The advantage of the proposed method is that the match process is performed simultaneously for all possible matching features. The simultaneous process allows a set of possible matching pairs to compete against other sets of possible matching pairs without putting any restrictions on the order of which a match must be performed. The method may be especially attractive when the geometric relationships between features in an image are unclear. We demonstrate the effectiveness of the proposed method using several examples.