Mexican hat wavelet

Mexican hat wavelet
Name	Mexican hat wavelet
Type	Continuous wavelet
Domain	Signal processing, harmonic analysis
Introduced	1980s (widespread use)
Related	Marr wavelet, Ricker wavelet, Morlet wavelet

Mexican hat wavelet

The Mexican hat wavelet is a real-valued, isotropic, second-derivative-of-Gaussian function used as a mother wavelet in continuous wavelet analysis and signal processing. It appears in literature on David Marr (neuroscientist), Georges-Michel Maréchal, and Lewis Fry Richardson-style multiscale modeling, and it has been applied in contexts ranging from Adrianovich-style geophysics to Richard Feynman-inspired intuition for multiresolution methods. The wavelet is closely associated with the Ricker wavelet used in seismology and with the Laplacian of Gaussian operator from image analysis pioneered by researchers such as Marr and Turing.

Definition and mathematical form

The Mexican hat wavelet ψ(x) is given by the negative normalized second derivative of a Gaussian: ψ(x) = (1 - x^2) e^{-x^2/2} (up to scale normalization), a form studied by Ronald A. Fisher-style statisticians and harmonic analysts like Andrey Kolmogorov. In multiple dimensions the isotropic form uses the Laplacian ∇^2 of the Gaussian e^{-||x||^2/2}, a construction related to techniques from S. R. Srinivasa Varadhan-influenced probability theory and Norbert Wiener-type stochastic analysis. Normalization constants are chosen to satisfy L^2 normalization conventions used by authors such as Ingrid Daubechies and Alexandre Grothendieck in functional analysis contexts.

Properties and admissibility

The Mexican hat is real, symmetric, and has zero mean, satisfying the admissibility condition required for continuous wavelet transforms as formalized by analysts like Yves Meyer and Stéphane Mallat. Its Fourier transform is expressible in closed form via a Gaussian multiplied by a polynomial, a property exploited by André Weil-style representation-theoretic techniques and by applied mathematicians such as John von Neumann for spectral considerations. The wavelet has compactly decaying tails (but is not compactly supported), a trade-off discussed in works by Paul Cohen and Elias Stein. Admissibility depends on nonvanishing integrability conditions similar to those in Calderón reproducing formulas used by A. P. Calderón.

Continuous and discrete transforms

The continuous wavelet transform (CWT) using the Mexican hat analyzes signals across scales and translations, following frameworks developed by Jean Morlet, Alex Grossmann, and Stephane Mallat. Discretization to a discrete wavelet transform (DWT) is nontrivial because the Mexican hat is not associated with an orthonormal multiresolution analysis like those constructed by Ingrid Daubechies; instead, discretized frames or redundant dictionaries derived from the Mexican hat are used in the tradition of Yves Meyer and David Donoho. Implementation of sampling, tiling of the time-scale plane, and reconstruction employ methods from numerical harmonic analysis advanced by Gilbert Strang and Eugenio Calabi-inspired finite-element thinking.

Relationships to other wavelets and functions

The Mexican hat is identical (up to scale) to the Ricker wavelet used in seismology, connecting it to applied work by Andrija Mohorovičić-inspired geophysicists and seismologists such as Inge Lehmann. It is also the Laplacian of the Gaussian (LoG), tying it to image-processing operators popularized by David Marr and earlier vision researchers like Hubel and Wiesel in biological vision. Comparisons are often drawn with the Morlet wavelet introduced by Jean Morlet and Alex Grossmann, and with compactly supported wavelets constructed by Ingrid Daubechies and orthogonal bases studied by Alfred Haar. Connections to spline-based constructions appear in work by Carl de Boor and approximation theorists like Isaac Schoenberg.

Applications

The Mexican hat wavelet has been used in edge detection and blob detection in image analysis following David Marr and Christopher Longuet-Higgins methods, in geophysical signal processing in the tradition of Charles Richter and Beno Gutenberg, and in neurophysiological modeling influenced by Hubel and Wiesel. It appears in astronomical data analysis used by teams similar to those in European Space Agency and National Aeronautics and Space Administration pipelines, in time-frequency analysis in seismology influenced by Robert H. Nott-style work, and in machine-learning feature extraction building on ideas from Yann LeCun and Geoffrey Hinton. The wavelet is also used in numerical solution of partial differential equations by researchers like John von Neumann and Sergiu Klainerman for multiscale preconditioning.

Numerical implementation and examples

Practical computation of Mexican hat transforms uses FFT-based convolution methods attributed to computational pioneers such as James Cooley and John Tukey; scale discretization and reconstruction employ frame theory developed by Ingrid Daubechies and Ronald Coifman. In image processing, discrete convolution kernels approximating the Laplacian of Gaussian are implemented in libraries inspired by work from Ken Thompson-era software and modern toolkits used at Los Alamos National Laboratory and Lawrence Berkeley National Laboratory. Example use cases include blob detection in microscopy data analyzed by groups around Erwin Neher and Bert Sakmann, and seismic wavelet matching in exploration geophysics practiced by firms with historical ties to Guglielmo Marconi-era signal work.

Category:Wavelets