finding height and width of an image using numpy and pixel locations

Question

I've converted two images into numpy arrays

image1Array

image2Array

Both images have been converted to grayscale, so there are only 0 or 255 values.

In both examples, there are many rows of white at the top (and bottom):

[255 255 255 255 .... 255 255 255 255]

I believe what i'm referring to as 'row' is really an array. I'm new to using Numpy. So, there is an array for every line in the image, and every pixel in that line is represented with a 0 or 255.

How do I find the first row that contains a black 0 pixel and the last row that contains a black 0 pixel? I should be able to use that to calculate the height. In these examples, this should be approximately the same number.

I believe numpy.where(image1Array == 0)[0] is returning the row of every black pixel; min() and max() of that seems to be what i'm looking for, but i'm not sure yet.

Conversely, how do I find the width of each image? In these examples, Image 2 should have a larger width number than Image 1

EDIT

I think all I need is something like this:

Height (the difference between the first row with a black pixel and the last row with a black pixel):

(max(numpy.where(image1Array == 0)[0])) - (min(numpy.where(image1Array == 0)[0]))

Width (the difference between the lowest column value with a black pixel and the highest column value with a black pixel):

(max(numpy.where(image1Array == 0)[1])) - (min(numpy.where(image1Array == 0)[1]))

So far, my testing is showing this is correct. Comparing the two images in the example above, their heights are equal while image2Array's width is double that of image1Array.

Answer 1

You would want something like this:

mask = x == 0  # or `x != 255` where x is your array

columns_indices = np.where(np.any(mask, axis=0))[0]
rows_indices = np.where(np.any(mask, axis=1))[0]

first_column_index, last_column_index = columns_indices[0], columns_indices[-1]
first_row_index, last_row_index = rows_indices[0], rows_indices[-1]

---

Explanation:

Let's create an example array using np.pad¹

import numpy as np


x = np.pad(array=np.zeros((3, 4)), 
           pad_width=((1, 2), (3, 4)),
           mode='constant',
           constant_values=255)
print(x)

[[255. 255. 255. 255. 255. 255. 255. 255. 255. 255. 255.]
 [255. 255. 255.   0.   0.   0.   0. 255. 255. 255. 255.]
 [255. 255. 255.   0.   0.   0.   0. 255. 255. 255. 255.]
 [255. 255. 255.   0.   0.   0.   0. 255. 255. 255. 255.]
 [255. 255. 255. 255. 255. 255. 255. 255. 255. 255. 255.]
 [255. 255. 255. 255. 255. 255. 255. 255. 255. 255. 255.]]

From here we can get a boolean mask array of zero elements as simple as that:

mask = x == 0
print(mask)

[[False False False False False False False False False False False]
 [False False False  True  True  True  True False False False False]
 [False False False  True  True  True  True False False False False]
 [False False False  True  True  True  True False False False False]
 [False False False False False False False False False False False]
 [False False False False False False False False False False False]]

Now we could use np.any to get those rows where there is at least one zero-element.
For columns:

print(np.any(mask, axis=0))
>>> [False False False  True  True  True  True False False False False]

and for rows:

print(np.any(mask, axis=1))
>>> [False  True  True  True False False]

Now we only have to convert boolean arrays to arrays of indices²:

columns_indices = np.where(np.any(mask, axis=0))[0]
print(columns_indices)
>>> [3 4 5 6]

rows_indices = np.where(np.any(mask, axis=1))[0]
print(rows_indices)
>>> [1 2 3]

Getting the first and the last rows/columns indices from here is pretty easy:

first_column_index, last_column_index = columns_indices[0], columns_indices[-1]
first_row_index, last_row_index = rows_indices[0], rows_indices[-1]

---

Timings:
I used the following code to calculate timings and plot them: Plot timings for a range of inputs.
Comparing my version with your version but refactored as this:

indices = np.where(x == 0)
first_row_index, last_row_index = indices[0][0], indices[0][-1]
first_column_index, last_column_index = indices[1][0], indices[1][-1]

plot_times([georgy_solution, yodish_solution],
           np.arange(10, 200, 5), 
           repeats=500)

plot_times([georgy_solution, yodish_solution],
           np.arange(200, 10000, 800), 
           repeats=1)

---

¹ See How to pad numpy array with zeros for nice examples.
² How to turn a boolean array into index array in numpy.

Answer 2

Here's a script that should work with python 2 or python 3. If you use IPython or jupyter, the "magic" %pylab does all the importing for you, and you don't need all the from... statements. After these statements, we create an image like the one you posted.

from __future__ import print_function # makes script work with python 2 and python 3
from matplotlib.pyplot import show, imshow, imread
from matplotlib.mlab import find
from numpy import zeros, int8, sum

img1 = zeros((256, 256), dtype=int8)
img1[50:200, 100:150] = 100
imshow(img1) 
show() # You don't need this call if you are using ipython or jupyter
# You now see a figure like the first one you posted
print('Axis 0 blob limits', find(sum(img1, axis=0) != 0)[[0, -1]]) 
print('Axis 1 blob limits', find(sum(img1, axis=1) != 0)[[0, -1]])

Using the sum function with an explicit specification for axis makes it return the sum along the given direction. the find function returns an array of all the indexes where the condition is true, in this case "where is the column or row sum zero?" Finally, slicing [0, -1] selects the first and last columns found.

If your image does not have any rows or any columns with all zeros, find returns an empty array, and the indexing attempt [0, -1] raises an IndexError. You can pretty the error condition up if you wrap a try...except block around it.