ashkarin/ndpatch

Package to extract arbitrary regions from an N-dimensional numpy array assuming it mirrored infinitely.

---

NDPatch is the package for extracting arbitrary regions from an N-dimensional numpy array assuming it mirrored infinitely.

Installation

The easiest way to install the latest version is by using pip:

$ pip install ndpatch

You may also use Git to clone the repository and install it manually:

$ git clone https://github.com/ashkarin/ndpatch.git
$ cd ndpatch
$ python setup.py install

Usage

To take a patch from the array:

import numpy as np
import ndpatch
array = np.arange(25).reshape((5,5))
index = (1, 2)
shape = (3, 3)
patch = ndpatch.get_ndpatch(array, shape, index)
# patch =
# [[ 7,  8,  9],
#  [12, 13, 14],
#  [17, 18, 19]]

To take get a random patch index:

import numpy as np
import ndpatch
array_shape = (5, 5)
index = ndpatch.get_random_patch_index(array_shape)

To extract random patches from the array:

import numpy as np
import ndpatch
npatches = 10
patch_shape = (3, 3)
array = np.arange(100).reshape((10,10))
patches = [ndpatch.get_random_ndpatch(array, patch_shape) for _ in range(npatches)]

To split the 3D array on set of overlapping 3D patches and rebuild it back:

import numpy as np
import ndpatch
array = np.arange(0, 125).reshape((5,5,5))
patch_shape = (4, 3, 3)
overlap = 2
indices = ndpatch.get_patches_indices(array.shape, patch_shape, overlap)
patches = [ndpatch.get_ndpatch(array, patch_shape, index) for index in indices]
reconstructed = ndpatch.reconstruct_from_patches(patches, indices, array.shape, default_value=0)
# Validate
equal = (reconstructed == array)
assert (np.all(equal))