Computationally efficient graph matching via energy vector extraction

Abstract

This paper presents a method for graph matching based on domain knowledge by quantifying representative graph features. Our method searches and extracts the most relevant cues in different graphs. Once these cues are extracted and quantified, a new energy function is used to match the different graphs based on the obtained features values. This approach has been successfully applied for deformable template matching. As a result the error of matching is reduced, as well as the computational cost by efficiently selecting and grouping representative features.