A Quantitative Evaluation of Symmetry Detection Algorithms

Symmetry is one of the most important cues for human and machine perception of the chaotic real world. For over three decades now, automatic symmetry detection from images/patterns has been a standing topic in computer vision. We observe a surge of new symmetry detection algorithms that go beyond simple bilateral symmetry detection. This paper presents a systematic, quantitative evaluation of rotation, reflection and translation symmetry detection algorithms published within the past 1.5 years. We provide a set of carefully chosen synthetic and real images that contain both single and multiple symmetries and a diverse range of computational challenges. We also provide their associated, hand-labeled ground truth. We propose a well-defined quantitative evaluation scheme for an effective validation and comparison of different symmetry detection algorithms. Our results indicate that even after several decades of effort, symmetry detection from real-world images remains a challenging, unsolved problem in computer vision.