Scale space contour approximation for shape similarity evaluation

Reliable and efficient comparison of two shape contours is presented. The shape contours are described by polygonal approximations derived using the linear diffusion equation solution. The vertices are extracted from the curve evolution generated in different scales, using the original contour as the initial condition. They are chosen between the points where curvature extremes exist, using a proper strategy. It results in a scalable representation of the shape contour that allows efficient and reliable computation of a similarity measure between two shapes. It is shown that this similarity measure is suitable for shape based retrieving. Key-words: Shape, Contour, Scale Space, Polygonal Approximation, Similarity Metric, Shape Retrieval.