Abstract: A new shape feature descriptor was proposed, which considers shape context features as an array of 1-D signals. The 1-D Fourier transformation was then performed on every signal. The new feature descriptor has been proved theoretically to be invariant for scaling, translation and rotation, with acceptable computational complexity. Shape point matching experiment shows the efficiency and effectiveness of the new feature descriptor.

Key words: 1-D Fourier transformation, shape context, rotation invariant