A Shape Representation Based on Polar Vector Fourier Descriptors

摘要

Fourier descriptors (FDs) are the widely used shape descriptors. But FDs can only be used to objects with single boundary, and it is inapplicable to objects with several components. In this paper, a region-based method, which is called Polar Vector Fourier Descriptors (PVFDs), is developed to extract invariant features. Firstly, the Cartesian coordinate system is converted into polar coordinate system. Secondly, the original two-dimensional image is reshaped into a vector (which is called Polar Vector (PV)) by linking radial vector end-to-end along the polar angle. Then, PVFDs are derived by applying Fourier transformation to PV. Consequently, the derived PVFDs are invariant to scaling and rotation. Furthermore, it is applicable to objects with several components (such as some Chinese characters etc.). Experimental results show that the proposed method is superior to Hu's moments and complex moments (CMs).