Symmetry-aware graph matching

• It is the first time symmetry constraints that are introduced into graph matching.
• The symmetry-aware graph matching improves the matching performance.
• A new dataset with annotation for evaluating graph matching is constructed.

Visual symmetry encodes middle- to high-level geometric information and plays an important role in many computer vision applications. Not much effort, however, has been devoted to utilize symmetry information for graph matching. In this paper, we propose a new framework for symmetry-aware graph matching in the context of image matching. In the framework, we first define symmetry matrices to characterize the geometric symmetry relation among image features, and then develop two methods to discover such symmetry. After that we design general strategies to boost graph matching algorithms with symmetry constraints, and apply these strategies to several state-of-the-art algorithms. For evaluation, the proposed symmetry-aware graph matching algorithms are first applied to two datasets involving clear symmetry patterns: a recently proposed architecture image set and a dental panoramic radiograph collection. The results clearly demonstrate the benefits of using symmetry information. An experiment is also conducted on a dataset where symmetry is less obvious, the results show that our strategies still achieve better of similar performance compared with state-of-the-arts.