/pyPyrTools

Python tools for multi-scale image processing, including Laplacian pyramids, Wavelets, and Steerable Pyramids http://www.cns.nyu.edu/~lcv/software.php

Primary LanguageJupyter NotebookMIT LicenseMIT

pyPyrTools

A Python 2.7 port of Eero Simoncelli's matlabPyrTools to Python. This port does not attept to recreate all of the matlab code from matlabPyrTools. The goal is to create a Python interface for the C code at the heart of matlabPyrTools, which implements flexible convolution with subsampling and multiple options for boundary-handling.

All code should be considered a beta release. Unit tests are available in TESTING/unitTests.py.

If you're using functions or parameters that do not have associated unit tests you should test this yourself to make sure the results are correct. You could then submit your test code, so that we can build more complete unit tests.

Authors

Rob Young and Eero Simoncelli, 7/13 William Broderick, 6/17

Installation

It's recommended you install from pip: pip install pyPyrTools. The pip install only been tested on Linux, there is no guarantee that it will work on other systems. Installing from source (see below) should work on OSX as well, but Windows is NOT supported.

NOTE: You will also need to install PyQt (either PyQt4 or PyQt5), but I have often been unable to install either from pip and so it is not listed as a required package for pip's setup.py. I would like eventually to remove this requirement, but for the time being you will have to install it on your own. See here for PyQt4 and here for PyQt5.

But if you want to install from source or pip doesn't work, you'll need to compile the associated C code yourself: assuming your path to python libraries is /usr/local/anaconda2, you would type:

gcc -shared -L/usr/local/anaconda2/lib -I/usr/local/anaconda2/include/python2.7/ -lpython2.7 -o wrapConv.so -fPIC convolve.c edges.c wrap.c internal_pointOp.c

Other requirements:

  • numpy
  • scipy
  • matplotlib
  • Pillow

Usage:

method parameters mimic the matlab function parameters except that there's no need to pass pyr or pind, since the pyPyrTools version pyr and pyrSize are properties of the class.

  • load modules (note that if you installed via pip, you can skip the first two lines):
>> import sys
>> sys.path.append('path to pyPyrTools parent directory')
>> import pyrPyrTools as ppt
  • create pyramid:
>> myPyr = ppt.Lpyr(img)
  • reconstruct image from pyramid:
>> reconImg = myPyr.reconLpyr()

Please see TUTORIALS/pyramids.ipynb for more examples. You can start this with: jupyter notebook pyramids.ipynb if you have iPython and Jupyter installed.

Testing

To use the included test, navigate to pyPyrTools/TESTS/ and run python unitTests.py. Note that a large amount of text will appear in your terminal window, even if the tests are succeeding, so don't freak out.

Function list

  • showIm(image, range, label, colormap, colorbar)

    display an image in a figure window

    • image - a 2D numpy array
    • range - a two element tuple. It specifies the values that map to the min and max colormap value. Passing a value of 'auto' (default) sets range=[min,max]. 'auto2' sets range=[mean-2*stdev, mean+2*stdev]. 'auto3' sets range=[p1-(p2-p1)/8, p2+(p2-p1)/8], where p1 is the 10th percentile value of the sorted matrix samples, and p2 is the 90th percentile value.
    • label - a string that is used as a figure title
    • colormap - either the string 'auto' (grey colormap will be used) or a string that is the name of a colormap variable
    • colorbar - a boolean that specifies whether or not a colorbar is displayed

  • maxPyrHt(imsz, filtsz)

    return the maximum possible pyramid height from the given image and filter size

    • imsz - integer giving the image size
    • filtsz - integer giving the filter size

  • binomialFilter(size)

    returns a numpy array of binomial coefficients of order (size-1)

    • size - integer giving size of filter

  • namedFilter(name)

    returns numpy array of named filter. Supported names are: binom<num> (where is a number denoting the size of the filter), 'qmf5', 'qmf9', 'qmf13', 'qmf8', 'qmf12', 'qmf16', 'haar', 'daub2', 'daub3', 'daub4', 'gauss5', 'gauss3'.

  • mkRamp(size, direction, slope, intercept, origin)

    Compute a matrix of dimension size (a (Y X) tuple, or a scalar (for square matrices)) containing samples of a ramp function, with a given gradient direction (radians, CW from X-axis, default=0), slope (per pixel, default = 1), and a value of intercept (default = 0) at the origin (default = (size+1)/2, (1 1) = upper left). All but the first argument are optional.

  • sp0Filters - steerable pyramid filters (see pyPyrUtils.py for references)

  • sp1Filters - steerable pyramid filters (see pyPyrUtils.py for references)

  • mkImpulse(size, origin, amplitude)

    Compute a matrix of dimension size (a (Y X) tuple, or a scalar (for square matrices)) containing a single non-zero entry, at position origin (defaults to ceil(size/2)), of value amplitude (defaults to 1).

  • Lpyr(image, height, filter1, filter2, edges)

    Laplacian pyramid: image parameter is required, others are optional

    • image - a 2D numpy array
    • height - an integer denoting number of pyramid levels desired. Defaults to maxPyrHt from pyPyrUtils.
    • filter1 - can be a string namimg a standard filter (from pyPyrUtils.namedFilter()), or a numpy array which will be used for (separable) convolution. Default is 'binom5'.
    • filter2 - specifies the "expansion" filter (default = filt1).
    • edges - specifies edge-handling. Options are:
      • 'circular' - circular convolution
      • 'reflect1' - reflect about the edge pixels
      • 'reflect2' - reflect, doubling the edge pixels
      • 'repeat' - repeat the edge pixels
      • 'zero' - assume values of zero outside image boundary
      • 'extend' - reflect and invert
      • 'dont-compute' - zero output when filter overhangs imput boundaries.

    Methods:

    • reconLpyr(levs, filt2, edges): Reconstruct image from Laplacian pyramid object
      • levs (optional) - numpy array of levels to include, or the string 'all' (default). The finest scale is 0.
      • filt2 (optional) - valid string name for pyPyrUtils.namedFilter() or a numpy array which will be used for (separable) convolution. Default = 'binom5'.
      • edges (optional) - same as edges for constructor above.
    • pyrLow(): Returns the coarsest band.
    • showPyr(range, gap, level_scale_factor): Show all bands of the pyramid in a figure window.
      • range - a two element tuple specifying the values that map to black and white respectively. These values are scaled by level_scale_factor**(lev-1) for bands at each level. Passing a value of 'auto1' sets range to the min and max values of matrix. 'auto2' sets range to 3 standard deviations below and above 0.0. In both of these cases, the lowpass band is independently scaled. A value of 'indep1' sets the range of each subband independently, as in a call to pyPyrUtils.showIm(subband, 'auto1'). Similarly, 'indep2' causes each subband to be scaled independently as if by pyPyrUtils.showIm(subband, 'indep2'). The default value for range is 'auto1' for 1D images and 'auto2' for 2D images.
      • gap - specifies the gap in pixels to leave between subbands (2D images only). default = 1.
      • level_scale_factor - indicates the relative scaling between pyramid levels. This should be set to the sum of the kernel taps of the lowpass filter used to construct the pyramid (default assumes L2-normalized filters, using a value of 2 for 2D images, sqrt(2) for 1D images).

  • Gpyr(image, height, filter, edges)

    Gaussian pyramid (subclass of Lpyr). image parameter is required, others are optional.

    • image - a 2D numpy array
    • height - an integer denoting number of pyramid levels desired. Defaults to maxPyrHt from pyPyrUtils.
    • filter - can be a string namimg a standard filter (from pyPyrUtils.namedFilter()), or a numpy array which will be used for (separable) convolution. Default is 'binom5'.
    • edges - specifies edge-handling. Options are:
      • 'circular' - circular convolution
      • 'reflect1' - reflect about the edge pixels
      • 'reflect2' - reflect, doubling the edge pixels
      • 'repeat' - repeat the edge pixels
      • 'zero' - assume values of zero outside image boundary
      • 'extend' - reflect and invert
      • 'dont-compute' - zero output when filter overhangs imput boundaries.

  • Spyr(image, height, filter, edges)

    Steerable pyramid. image parameter is required, others are optional

    • image - a 2D numpy array

    • height - an integer denoting number of pyramid levels desired. Defaults to maxPyrHt from pyPyrUtils. You can specify 'auto' to use this value.

    • filter - The name of one of the steerable pyramid filters in pyPyrUtils: 'sp0Filters', 'sp1Filters', 'sp3Filters', 'sp5Filters'. Default is 'sp1Filters'.

    • edges - specifies edge-handling. Options are:

      • 'circular' - circular convolution
      • 'reflect1' - reflect about the edge pixels
      • 'reflect2' - reflect, doubling the edge pixels
      • 'repeat' - repeat the edge pixels
      • 'zero' - assume values of zero outside image boundary
      • 'extend' - reflect and invert
      • 'dont-compute' - zero output when filter overhangs input boundaries.

    • spyrHt() - return the height of the pyramid

    • numBands() - return the number of bands in the pyramid

    • pyrLow() - return the lowest band

    • pyrHight() - return the highest band

    • reconSpyr(filter, edges, levs, bands) - reconstruct image from pyramid

      • filter - (optional) same as for constructor above. Default is 'sp1Filters'.
      • edges - (optional) same as for constructor above. Default is 'reflect1'.
      • levs - (optional) should be a numpy array of levels to include, or the string 'all' (default). 0 corresponds to the residual highpass subband. 1 corresponds to number spyrHt()+1.
      • bands - (optional) should be a list of bands to include, or the string 'all' (default). 0 = vertical, rest proceeding anti-clockwise.