Empirical Wavelet Transforms
The Empirical Wavelet Transform (EWT) aims to decompose a signal or an image on
wavelet tight frames which are built adaptively. In 1D, the procedure consists in
detecting the supports of some "modes" in the Fourier spectrum and then using these supports
to build Littlewood-Paley type wavelets. In 2D, based on the same principle, we propose
empirical versions of the tensor wavelet transform, a 2D Littlewood-Paley transform,
the Ridgelet transform and the Curvelet transform. The advantage of this empirical
approach is to keep together some information that otherwise would be split in the case of
dyadic filters. The provided Matlab toolbox performs all these transforms.
You can refer to these papers to learn more about these wavelets:
1D
and
2D
Feel free to contact me if you need some help, report a bug, or any
other suggestions.
Some examples of EWT coefficients in 1D and 2D:
- 1D EWT on an Electrocardiogram signal:
|
|
ECG signal |
EWT components |
- 2D Empirical Curvelet Transform - II:
|
|
Input image |
Detected Fourier supports |
The Empirical Wavelet Transform Matlab Toolbox Archive
The current version is
- v3.0 (July 21, 2015) bug fixes in the curvelet transform (crash due to odd image sizes
+ wrong high frequency filters for option 2) + new curvelet transform option 3 (scales per angular sectors)
- v2.0 (April 24, 2014) ftc algorithm removed + bugs fixes + new functions
- v1.2 (June 12, 2013) ftc_seg.c fixed and can now be compiled on Windows
- v1.0 (June 10, 2013)
Toolbox available at
Matlab
Central.