A feature of 2D and 3D images invariant to similarity transformations based on the bispectrum
Yo Horikawa · Systems and Computers in Japan · 2002
Abstract A novel feature of 2D and 3D patterns invariant to similarity transformations is derived from the bispectrum. The invariant feature is a histogram of triple products of frequency components and is calculated with the Fourier transform of original images. Computer experiments on sinusoidal patterns, 2D texture images, and 3D brain images show that the invariant feature is applicable to the classification of images shifted and rotated arbitrarily, scaled up to double size, and including additive noise up to 0 dB. © 2002 Scripta Technica, Syst Comp Jpn, 33(3): 1–10, 2002