Detection of Mirror-Symmetric Image Patches

Viorica Patraucean, Rafael Grompone von Gioi, Maks Ovsjanikov

Abstract: We propose a novel approach for detecting partial reflectional symmetry in images. Our method consists of two principal stages: candidate selection and validation. In the first step, candidates for mirror-symmetric patches are identified using an existing heuristic procedure based on Hough voting. The candidates are then validated using a principled statistical procedure inspired from the a contrario theory, which minimizes the number of false positives. Our algorithm uses integral image properties to enhance the execution time.