Steerable Pyramid

Python3 Implementation of steerable pyramid.

This is based on the article bellow.

The Heeger-Bergen Pyramid-Based Texture Synthesis Algorithm, Briand,T. et al. (2014)

According to this article, this is a "real version" of steerable pyramid which is used in Portilla and Simoncelli(2000).

Parametric Texture Model Based on Joint Statistics of Complex Wavelet Coefficient, Portilla, J. and Simoncelli, E.(2000)

Results

• Depth : 4
• Orientation : 4 (0, pi/4, pi/2, 3*pi/4)

Original Image

Pyramid Images

band frequency

  

low frequency residual

high frequency residual

Reconstructed Images

Images of Filters (Fourier Domain)

First Highpass Filter (H0 in Portilla and Simoncelli)

First Lowpass Filter (L0 in Portilla and Simoncelli)

Bandpass Filters (B*L in Portilla and Simoncelli)

Usage

Environment

• python3.5 (3.0+)
• GPU is not used.

Execution

• create 'out' directory.
• python create_collapse_pyramid.py -i saucer-mono256.png -x 256 -y 256, if you want to know details about parameters, see source code

References

• The Heeger-Bergen Pyramid-Based Texture Synthesis Algorithm, Briend, T. et al.(2014)
• Parametric Texture Model Based on Joint Statistics of Complex Wavelet Coefficient, Portilla, J. and Simoncelli, E.(2000)
• The Steerable Pyramid:A Flexible Architecture For Multi-Scale Derivative Computation, Simoncelli,E. and Freeman, W.(1995)
• A Filter Design Technique For Steerable Pyramid Image Transform, Karasaridis, A. and Simoncelli,E.(1996)
• Design and Use of Steerable Filters, Freeman,W. and Adelson,E.(1991)