Spiral Symmetry Detection Based On Robust Feature Extraction And Matching

Symmetry is ubiquitous in real world and symmetry detection is a hot topic in computer vision. This paper focuses on rotational symmetry detection and creatively proposes the problem of infinite rotational symmetry detection, i.e., spiral symmetry detection. Although spiral symmetry is omnipresent, none of the existing methods have been devoted to solving single unsegmented image spiral symmetry structure detection.To solve this challenging problem, we propose a novel spiral symmetry detection and reconstruct method which can stably detect the potential spiral symmetry centroids in a real world image, correctly judge the spiral growing direction, accurately estimate the spiral growth rate, e?ectively obtain the evident initial radii set, and finally get the spiral symmetry reconstruction result.The innovativeness of our work mainly includes four aspects. Firstly, we generalize the concept of rotational symmetry detection from radial symmetry detection to spiral symmetry detection and propose a matched feature triplet(MFT) extraction method as the basement of our spiral symmetry detection approach. Secondly, we propose an MFT based weighted Hough space voting strategy to realise spiral symmetry centroids detection. Thirdly, we propose a significance weights accumulation and decision method based on one-loop property of MFT to determine the spiral growing direction, a weighted Gaussian-blurred Hough space voting strategy to estimate the spiral growth rate, and a global radial reconstruction error minimization algorithm to get the evident spiral fragments in a spiral symmetry structure. Finally, we realise the tropical cyclone(TC) spiral symmetry region detection and reconstruction on single unsegmented TC image using the proposed method.