The Hermite Transformation in Quaternionic Analysis

The conventional way of describing an image is in terms of its canonical pixel-based representation. Other image description techniques are based on image transformations. Such an image transformation converts a canonical image representation into a representation in which specific properties of an image are described more explicitly. In most transformations, images are locally approximated within a window by a linear combination of a number of a priori selected patterns. The coefficients of such a decomposition then provide the desired image representation. The Hermite transform is an image transformation technique introduced by Martens. It uses overlapping Gaussian windows and projects images locally onto a basis of orthogonal polynomials. As the analysis filters needed for the Hermite transform are derivatives of Gaussians, Hermite analysis is in close agreement with the information analysis carried out by the human visual system. In this paper we construct a new higher dimensional Hermite transform within the framework of Quaternionic Analysis. The building blocks for this construction are the Clifford-Hermite polynomials rewritten in terms of Quaternionic analysis. Furthermore, we compare this newly introduced Hermite transform with the Quaternionic-Hermite Continuous Wavelet transform. The Continuous Wavelet transform is a signal analysis technique suitable for non-stationary, inhomogeneous signals for which Fourier analysis is inadequate. Finally the developed three dimensional filter functions of the Quaternionic-Hermite transform are tested with traditional scalar benchmark signals upon their selectivity at detecting pointwise singularities