3D point set registration of Chinese calligraphy

Abstract

This paper presents an innovative method to align two handwritten Chinese characters modeled by three-dimensional point sets. A piecewise cubic Bezier curve is first defined for each character stroke, interpolating the 2D sample points. The density of sample points is then adjusted along the curve for even distribution. Later all 2D points are extended into 3D space to integrate additional geometry information. Registration is then performed between two 3D point sets using Gaussian Mixture Model (GMM). This paper establishes the registration by minimizing the Euclidean Distance(L2) between two GMMs and then apply transformation accordingly. The key idea of this process is to include extra geometric information of the calligraphy to facilitate the registration process, using simple algorithms. Through the alignment and transformation process, the global layout of character is adjusted, but with local features retained. This technique can be applied in the morphing and beautification of Chinese calligraphy.