Symmetry detection is an important task that arises in MRI analysis, TEM microscopy and other scientific and industrial problems. Current methods of symmetry detection usually analyse SIFT invariants or use voting schemas. The article proposes and analyses an optimization problem that can be used to describe the image symmetry detection pipeline in terms of correlation and similarity. The article also describes an algorithm that solves the aforementioned optimization problem using a ransom sample consensus voting scheme, thus bringing the proposed method closer to other RANSAC-based methods. It is shown that an implementation based on the proposed model outperforms SIFT-based detectors, and stays in row with other voting-based methods; it is also shown that the algorithm's parameters could be adjusted for specific tasks to improve the detection quality. The article also compares the proposed method with other existing solutions in terms of accuracy and precision; visualizations are also included.

Keywords: RANSAC, random sample consensus, reflection symmetry, computer vision, image processing