Detecting Rotational Symmetry via Global/Local Image Analysis

Yousuke Inagaki, Kanji Tanaka
2013 IAPR International Workshop on Machine Vision Applications  
We propose a unified framework for rotational symmetry detection that combines the advantages of global and local image analysis. While existing methods based on global image analysis have proven to achieve stateof-the-art results, potential symmetries detected by the global methods still involve many false positives. Our idea is to employ efficient local image analysis as a cue, to intelligently choose most likely symmetry candidates, out of the potential candidates detected by the global
more » ... d. We cast the sub-task of local image analysis as a correspondence growing task and introduce a robust solution employing an MCMC sampling algorithm. We demonstrate improvements over the best known method on challenging real images.
dblp:conf/mva/InagakiT13 fatcat:ge7er3lncjez5bmvgnzdhebxgm