PyWavelets

1.9.0 · active · verified Sun Apr 05

PyWavelets is a free, open-source Python library for wavelet transforms. It provides 1D, 2D, and nD forward and inverse Discrete Wavelet Transforms (DWT and IDWT), Stationary Wavelet Transforms (SWT), Wavelet Packet decomposition, and Continuous Wavelet Transforms (CWT). It combines a simple high-level interface with low-level C and Cython performance and is widely used in signal processing, image compression, and noise removal. The current version is 1.9.0 and it has an active release cadence, with minor updates and new features being released regularly.

Warnings

breaking PyWavelets 1.9.0 dropped support for Python 3.10 and added support for Python 3.14. Ensure your Python environment meets the minimum requirement of Python >=3.11.
Fix: Upgrade Python to 3.11 or newer, or downgrade PyWavelets to a compatible version if unable to upgrade Python.
breaking The minimum supported NumPy version has been updated. PyWavelets 1.9.0 explicitly requires NumPy >= 1.25.0. Using an older NumPy version will likely lead to installation or runtime errors.
Fix: Update NumPy to version 1.25.0 or higher using `pip install --upgrade numpy`.
breaking The optional dependency on SciPy for FFT operations was removed in PyWavelets 1.9.0. If your application implicitly relied on SciPy being available due to PyWavelets, it might now be missing.
Fix: Manually install `scipy` (e.g., `pip install scipy`) if your application or specific PyWavelets functions (like certain CWT functionalities) require it.
gotcha The image returned by `pywt.data.camera` was replaced in version 1.2.0 with a new CC0-licensed image due to licensing restrictions on the original 'cameraman' image. If your application relies on the exact pixel data of the original image, it will be different in newer versions.
Fix: For commercial use, the new image is safe. If the original image is critical for non-commercial use, you must source it externally or use an older PyWavelets version.
gotcha The length of coefficient arrays returned by `pywt.dwt` depends on the chosen signal extension `mode`. For all modes except 'periodization' ('per'), `len(cA) == len(cD) == floor((len(data) + wavelet.dec_len - 1) / 2)`. For 'periodization' mode, `len(cA) == len(cD) == ceil(len(data) / 2)`. This can lead to unexpected output sizes.
Fix: Always explicitly specify the `mode` parameter (e.g., `mode='symmetric'`) and consult the documentation or `pywt.dwt_coeff_len()` to predict output sizes, especially when chaining transforms or performing inverse transforms.

Install

pip install PyWavelets PyPI

Imports

pywt
```
import pywt
```

Quickstart

This quickstart demonstrates a basic 1D Discrete Wavelet Transform (DWT) and its inverse using the 'db1' (Daubechies 1) wavelet. It shows how to decompose a signal into approximation (cA) and detail (cD) coefficients, and then reconstruct the original signal.

import pywt
import numpy as np

data = np.array([1, 2, 3, 4, 5, 6])
wavelet = 'db1'

# Perform a single-level Discrete Wavelet Transform (DWT)
cA, cD = pywt.dwt(data, wavelet)

print(f"Original data: {data}")
print(f"Approximation coefficients (cA): {cA}")
print(f"Detail coefficients (cD): {cD}")

# Perform inverse DWT to reconstruct the signal
reconstructed_data = pywt.idwt(cA, cD, wavelet)
print(f"Reconstructed data: {reconstructed_data}")

view raw JSON →