Robust Feature Matching Through Maximal Cliques

Publisher Website
Google Scholar
Abstract
A crucial step in the recognition or location of an object in an image is the proper identification of object features. If the features are not uniquely characterized by their local appearances, as is often the case in programmable assembly, the matching technique must base its decisions on the relative structure of the features. In this paper we describe a technique that uses the relative positions and orientations of the features to determine the correspondence between features of an object model and features observed in a picture. A graph is constructed in which maximal cliques (i.e., completely connected subgraphs) represent mutually consistent assignments of model features to observed features. The technique is a robust, general-purpose way to match structures. However, in practical applications its use is restricted to moderately sized graphs because the algorithm that locates maximal cliques is apparently exponential. For tasks that require the analysis of large graphs a few techniques are presented to reformulate them so that smaller graphs are sufficient.

This publication has 0 references indexed in Scilit: