Wavelet transform

Wavelet transform
Alessio Damato · CC BY-SA 3.0 · source
Name	Wavelet transform
Introduced	1980s
Inventor	Jean Morlet; Ingrid Daubechies
Field	Signal processing; Applied mathematics
Related	Fourier transform; Short-time Fourier transform; Multiresolution analysis

Wavelet transform is a mathematical technique for decomposing functions, signals, or images into localized basis functions called wavelets. It provides time-frequency or space-scale representations useful in signal processing, image analysis, numerical analysis, and data compression. Developed through contributions by engineers and mathematicians, the transform links to multiresolution concepts that complement traditional Fourier-based methods.

Overview

The development of the method involved contributions from Jean Morlet, Alex Grossmann, Yves Meyer, Ingrid Daubechies, and researchers at institutions such as Bell Laboratories, École Polytechnique, and Columbia University. Early applications were driven by needs in geophysics at companies like TotalEnergies and by meteorological research at Météo-France. Subsequent work connected to fields represented by Society for Industrial and Applied Mathematics, IEEE, Association for Computing Machinery, Los Alamos National Laboratory, and European Space Agency. Key milestones include the construction of compactly supported orthonormal bases by Ingrid Daubechies and the formulation of multiresolution analysis by Stéphane Mallat. Influential publications appeared in journals affiliated with American Mathematical Society and conferences such as International Conference on Acoustics, Speech, and Signal Processing.

Mathematical foundations

At its core, the theory builds on concepts introduced by Joseph Fourier and extended by researchers at University of Cambridge and Massachusetts Institute of Technology. Multiresolution analysis formalized by Stéphane Mallat and constructions by Yves Meyer rely on function spaces studied by John von Neumann and Stefan Banach. Wavelet bases are formed via dilation and translation operators connected to representations studied in works associated with École Normale Supérieure and Institute for Advanced Study. Orthogonality, compact support, vanishing moments, and regularity tie to theorems developed by Ingrid Daubechies and analyses influenced by Norbert Wiener and Andrey Kolmogorov. Discrete formulations relate to filter bank theory advanced at Bell Laboratories and to sampling theory researched at Claude Shannon and Harry Nyquist. Continuous formulations use integrals akin to those in work by Henri Lebesgue and Sofia Kovalevskaya, while fast algorithms exploit algebraic structures examined by Emmy Noether and André Weil.

Types of wavelet transforms

The field distinguishes multiple forms, each studied in contexts involving universities and laboratories: continuous wavelet transform forms used in geophysics at Cornell University and Princeton University; discrete wavelet transform variants developed at Bell Laboratories and Darmstadt University of Technology; maximal overlap discrete wavelet transform analyses employed at NASA centers; complex wavelet transforms researched at University of British Columbia and University of Illinois Urbana-Champaign; and stationary wavelet transforms applied in projects at Lawrence Berkeley National Laboratory. Other specialized constructions include biorthogonal wavelets associated with École Polytechnique Fédérale de Lausanne and lifting schemes introduced by researchers at University of Cambridge. Compactly supported orthonormal wavelets by Ingrid Daubechies contrast with spline wavelets derived in work related to University of Washington. Continuous, discrete, redundant, and critically sampled families have been studied at institutions like ETH Zurich, University of California, Berkeley, and University of Oxford.

Implementation and algorithms

Efficient implementations draw on the fast wavelet transform and filter banks developed at Bell Laboratories and algorithmic techniques present in software from MathWorks and open-source projects at National Institute of Standards and Technology. Multiresolution decomposition algorithms influenced work at Sandia National Laboratories and within standards committees at International Telecommunication Union. Numerical stability and boundary handling reference methods tested by research groups at California Institute of Technology and Massachusetts General Hospital. Implementations use discrete filters, lifting schemes, and convolution routines optimized in toolkits from Google Research and Microsoft Research; parallel and GPU-accelerated versions were developed at Oak Ridge National Laboratory and Argonne National Laboratory. Formal verification of transforms has been pursued in collaborations with Carnegie Mellon University and University of Toronto.

Applications

Wavelet-based techniques have been applied across many domains: image compression standards in projects by Joint Photographic Experts Group and video codecs researched at Motion Picture Experts Group; denoising methods used in medical imaging at Mayo Clinic and Johns Hopkins University; seismic analysis conducted by US Geological Survey and Chevron; feature extraction in machine learning efforts at Google and Facebook; financial time-series analysis performed by teams at Goldman Sachs and Morgan Stanley; fault detection in engineering studied at General Electric and Siemens; biomedical signal processing at Harvard Medical School and Stanford University; climate data analysis at National Oceanic and Atmospheric Administration and Met Office; and astronomy applications in projects led by European Southern Observatory and National Aeronautics and Space Administration. Wavelets underpin compression in standards developed by International Organization for Standardization and contribute to algorithms in software from Adobe Systems.

Limitations and challenges

Challenges include selection of appropriate bases—a problem investigated in research groups at Princeton University and University of Chicago—and managing edge effects studied at Columbia University and Imperial College London. Trade-offs between redundancy and computational cost were explored at Bell Laboratories and IBM Research; handling nonstationary noise and adaptive schemes remains an active topic at National Institutes of Health and Wellcome Trust. Scalability for high-dimensional data and integration with deep learning architectures are ongoing research areas at DeepMind and OpenAI. Theoretical limits connected to time-frequency localization echo earlier debates involving Joseph Fourier and later mathematical formalizations by Hermann Weyl and André Weil.

Category:Signal processing