filterbanks: Filter Banks and Multiresolution Transform Layers

filterbanks provides TensorFlow/Keras layers for multiresolution and geometric filter banks.

Capabilities

• DCT2D / IDCT2D
• MultiLevelLiftingDWT1D_K
• MultiLevelLiftingDWT2D_K
• MultiLevelLiftingDWT3D_K
• LapPyrAnalysis / LapPyrSynthesis
• MonogenicPyramid2D
• SteerablePyramid2D
• Scattering2D
• WaveAtoms2DAnalysis / WaveAtoms2DSynthesis
• WignerVille1D
• filterbanks.tileviewer for tiling plots

Limitations

• Some layers are analysis-only and do not include a synthesis inverse.
• Exact perfect reconstruction applies only to paired forward/inverse classes where provided.
• Optional viewer and probe utilities may require extra packages such as matplotlib.

Installation

pip install filterbanks

Transform References

Package-Specific Layers

Minimal Example

import tensorflow as tf
from filterbanks import MultiLevelLiftingDWT2D_K

x = tf.random.normal([1, 64, 64, 1])
layer = MultiLevelLiftingDWT2D_K(levels=2, preset="haar")

pyr = layer(x)
y = layer.synth(*pyr)

print(x.shape, y.shape)

Tiling Viewer

python -m filterbanks.tileviewer.main_viewer --transform contourlet --levels 3 --Ks 4 8 8 --kappas 64 64 64 --H 256 --W 256 --save Figs/contourlet_tiling.png

Citation

This software is released for broad research, educational, and engineering use. If this package helps your work, please cite the following paper:

@misc{tarafdar2026interpretablefrugallearningsystems,
      title={Interpretable and Frugal Learning Systems Employing Multiresolution Pyramids and Volterra Kernels},
      author={Kishore Kumar Tarafdar},
      year={2026},
      eprint={2606.15011},
      archivePrefix={arXiv},
      primaryClass={eess.SP},
      url={https://arxiv.org/abs/2606.15011},
}

License

Apache License 2.0. See LICENSE.