Shape Similarity by Piecewise Linear Alignment
Xénophon, Trahanias, E Panos, Orphanoudakis, Jon Sporring, Xenophon Zabulis, Panos E. Trahanias, Stelios C. Orphanoudakis · 2016
A key problem when comparing planar shapes is to locate corresponding reference points, such as inflection or high curvature points. High curvature points are to be preferred, since they are psychophysically more important and are shown to be computationally more reliable. Using a scale-space of curves such ‘corners’ are located, and an empirical analysis demonstrates that high curvature points are indeed least sensitive to noise. Using reliable ‘corners’, a shift invariant and piecewise linear alignment of shapes is defined, and a similarity measure between two curves is defined using lossless compression. The measure is chosen to be proportional to the area between two shapes after piecewise alignment and to the number of data points. The proposed shape similarity is validated using a small shape database.