Circular-Fourier–radial-Mellin transform descriptors for pattern recognition

Abstract

Image descriptors based on a circular-Fourier-radial-Mellin transform are proposed. They are invariant with respect to rotation, translation, and change of scale. They represent a generalized approach to specific descriptors using a circular-harmonic expansion, a Mellin transform, or moment invariants. The possibility of computing the Fourier-Mellin descriptors by using an optical processor in real time is discussed.

Automatic voice onset time detection for unvoiced stops (/p/,/t/,/k/) with application to accent classification

“…The Mellin Fourier transform has been applied for image recognition because its resulting spectrum is invariant in rotation, translation, and scale (Chen et al, 1994;Sheng and Arsenault, 1986;Sheng and Duvernoy, 1986;Sheng et al, 1988). In the case of stop-analysis, as mentioned earlier, VOR has abrupt and short duration temporal characteristic changes together with the variation of the entire length, and we aim to separate the impact on accent classification based on duration as represented by VOT versus discriminating changes in spectral structure.…”

Spectral minutiae: A fixed-length representation of a minutiae set

“…Let T k (m, n) be defined as T (m, n) with a circular shift k in the horizontal direction. For each shift trial, a new similarity score S (R,T k ) is calculated using (14) or (16). Finally, the highest score is chosen as the final matching score and the corresponding shift k is recorded as the best shift (that is, the best rotation).…”

“…It was often used in image processing to obtain a translation, rotation and scaling invariant descriptor of the image [8,14]. However, the implementation of the Fourier-Mellin transform requires a Fourier transform and a polar-logarithmic mapping.…”

Robust and Efficient Fourier–Mellin Transform Approximations for Gray-Level Image Reconstruction and Complete Invariant Description

“…Therefore, the corresponding set of invariant features is dramatically reduced and only represents an outline of the shape of the object. For pattern recognition purposes, the classification of an unknown object as one of a set of reference patterns is achieved with a comparison method such as the computation of an error between the features [32], neural networks [33,34], or statistical classifiers [19,20,35]. These methods turn out to be efficient when the models to compare have simple and distinct shapes (typographic symbols or letters).…”