Disk-harmonic coefficients for invariant pattern recognition

Disk harmonics are defined, and an explanation is given as to why they can be interpreted as the natural generalization of the Fourier basis set onto the unit disk. An existing statistical theory is used to show that a particular set of disk-harmonic coefficients is more suitable for describing images on the unit disk than Zernike or pseudo-Zernike moments are. Zernike moments have been applied to a wide range of problems. However, we concentrate on the problem of invariant pattern recognition and briefly indicate other problems where disk-harmonic coefficients are possibly useful. The effects that different pixel resolutions and discrete white noise have on the three moment or coefficient sets compared in this paper are briefly investigated experimentally.