Rotation Invariant Shape Contexts based on Feature-space Fourier Transformation

摘要

We propose a new pixel-level shape descriptor. First, shape contexts are computed. Then, 2D FFT is performed on each 2D histogram from shape contexts. Such a scheme solves the rotation-invariance problem of shape contexts based on the shift theorem of Fourier Transformation while does not increase the computational complexity. Theoretical proof and experimental validation are provided.