Robust detection of skewed symmetries by combining local and semi-local affine invariants

Affine-invariant feature vector (Ip and Shen Image Vision Comput. 16 (2) (1998) 135-146), that captures both local and semi-local geometric features around each point of the object boundary is applied here for the detection of skewed symmetries. Based on the affine-invariant shape representation, the problem of detecting symmetry axes has been formulated as a problem of detecting lines, with known orientations, in a local similarity matrix of an object. Since the feature vector extracts sufficient local and semi-local shape information for every point along the object boundary, the process of checking symmetric point pairs is thus robust against both noises and deformations. Moreover, our technique is able to detect all the local reflectional symmetries contained in the object. Various experimental results have shown the robustness and effectiveness of our method in detecting skewed symmetries from both self-symmetric objects and generalized objects. © 2001 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.