Counting of adjacent cells in a numpy array

Question

Past midnight and maybe someone has an idea how to tackle a problem of mine. I want to count the number of adjacent cells (which means the number of array fields with other values eg. zeroes in the vicinity of array values) as sum for each valid value!.

Example:

import numpy, scipy
s = ndimage.generate_binary_structure(2,2) # Structure can vary
a = numpy.zeros((6,6), dtype=numpy.int) # Example array
a[2:4, 2:4] = 1;a[2,4] = 1 # with example value structure
print a 
>[[0 0 0 0 0 0]
  [0 0 0 0 0 0]
  [0 0 1 1 1 0]
  [0 0 1 1 0 0]
  [0 0 0 0 0 0]
  [0 0 0 0 0 0]]
# The value at position [2,4] is surrounded by 6 zeros, while the one at
# position [2,2] has 5 zeros in the vicinity if 's' is the assumed binary structure. 
# Total sum of surrounding zeroes is therefore sum(5+4+6+4+5) == 24

How can i count the number of zeroes in such way if the structure of my values vary? I somehow believe to must take use of the binary_dilation function of SciPy, which is able to enlarge the value structure, but simple counting of overlaps can't lead me to the correct sum or does it?

print ndimage.binary_dilation(a,s).astype(a.dtype)
[[0 0 0 0 0 0]
 [0 1 1 1 1 1]
 [0 1 1 1 1 1]
 [0 1 1 1 1 1]
 [0 1 1 1 1 0]
 [0 0 0 0 0 0]]

I'm not entirely clear on what you want here. So given a position, you want to know the number of neighbors whose value == 0? If you know the shape of the array (to find discover the edges) you can use simple math... annoying, time consuming (to write and process), but pretty straight forward. — monkut
– monkut, Commented Sep 27, 2012 at 2:16

nneonneo · Accepted Answer · 2012-09-27 03:14:50Z

9

Use a convolution to count neighbours:

import numpy
import scipy.signal

a = numpy.zeros((6,6), dtype=numpy.int) # Example array
a[2:4, 2:4] = 1;a[2,4] = 1 # with example value structure

b = 1-a
c = scipy.signal.convolve2d(b, numpy.ones((3,3)), mode='same')

print numpy.sum(c * a)

b = 1-a allows us to count each zero while ignoring the ones.

We convolve with a 3x3 all-ones kernel, which sets each element to the sum of it and its 8 neighbouring values (other kernels are possible, such as the + kernel for only orthogonally adjacent values). With these summed values, we mask off the zeros in the original input (since we don't care about their neighbours), and sum over the whole array.

edited Sep 27, 2012 at 3:14

answered Sep 27, 2012 at 3:09

nneonneo

181k37 gold badges331 silver badges412 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Warren Weckesser Over a year ago

I was just about to post almost the same answer: (a * ndimage.convolve((a == 0).astype(int), s)).sum()

Curlew Over a year ago

yeah, that exactly want i wanted. Great!

swang · Accepted Answer · 2012-09-27 02:21:55Z

1

I think you already got it. after dilation, the number of 1 is 19, minus 5 of the starting shape, you have 14. which is the number of zeros surrounding your shape. Your total of 24 has overlaps.

answered Sep 27, 2012 at 2:21

swang

2191 silver badge3 bronze badges

2 Comments

nneonneo Over a year ago

This can't be the right answer. Consider a 1 surrounded by 8 zeros. The correct answer is 8, but you will see 9 1s in the output, and your approach would add one to give 10.

nneonneo Over a year ago

Oh, sorry. You were counting surrounding zeros without overlaps. I think the poster wanted overlaps. But your solution is right for counting without overlap.

Collectives™ on Stack Overflow

Counting of adjacent cells in a numpy array

2 Answers 2

2 Comments

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related