get non zero ROI from numpy array

Question

I want to extract a rectangular ROI from an image. The image contains a single connected non zero part.

I need it to be efficient in run time.

I was thinking maybe:

1. Summing along each direction.
2. Finding first non zero and last non zero.
3. Slicing the image accordingly.

Is there a better way?

My code:

First is a function to find the first and last non zero:

import numpy as np

from PIL import Image

def first_last_nonzero(boolean_vector):
    first = last = -1
    for idx,val in enumerate(boolean_vector):
        if val == True and first == -1:
            first = idx
        if val == False and first != -1:
            last = idx
            return first , last

Then creating an image:

np_im = np.array([[  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0 255 154 251  60   0   0   0]
 [  0   0   0   0   4  66   0   0 255   0   0   0]
 [  0   0   0   0   0   0   0 134  48   0   0   0]
 [  0   0   0   0   0   0 236  70   0   0   0   0]
 [  0   0   0   0   1 255   0   0   0   0   0   0]
 [  0   0   0   0 255  24  24  24   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0]])

Then running our function on the sum along each axis:

y_start, y_end = first_last_nonzero(np.sum(np_im, 1)>0)
x_start, x_end = first_last_nonzero(np.sum(np_im, 0)>0)
cropped_np_im = np_im[y_start:y_end, x_start:x_end]

# show the cropped image
Image.fromarray(cropped_np_im).show()

This works but there are probably a plenty of unnecessary calculations. Is there a better way to do this? Or maybe more pythonic way?

Answer 1

You can make use of the functions from this post: Numpy: How to find first non-zero value in every column of a numpy array?

def first_nonzero(arr, axis, invalid_val=-1):
    mask = arr!=0
    return np.where(mask.any(axis=axis), mask.argmax(axis=axis), invalid_val)

def last_nonzero(arr, axis, invalid_val=-1):
    mask = arr!=0
    val = arr.shape[axis] - np.flip(mask, axis=axis).argmax(axis=axis) - 1
    return np.where(mask.any(axis=axis), val, invalid_val)


arr = np.array([
    [0, 0, 0, 0, 1, 1],
    [0, 0, 1, 1, 0, 0],
    [0, 1, 0, 0, 0, 0],
    [0, 0, 0, 1, 0, 0],
    [0, 0, 1, 0, 1, 0],
    [0, 0, 0, 0, 0, 0] ])

y_Min, y_Max, x_Min, x_Max = (0, 0, 0, 0) 
y_Min = first_nonzero(arr, axis = 0, invalid_val = -1) 
y_Min = (y_Min[y_Min >= 0]).min() 
x_Min = first_nonzero(arr, axis = 1, invalid_val = -1) 
x_Min = (x_Min[x_Min >= 0]).min() 
y_Max = last_nonzero(arr, axis = 0, invalid_val = -1) 
y_Max = (y_Max[y_Max >= 0]).max() 
x_Max = last_nonzero(arr, axis = 1, invalid_val = -1) 
x_Max = (x_Max[x_Max >= 0]).max() 
print(x_Min) 
print(y_Min) 
print(x_Max) 
print(y_Max)

For this example of mine, the code will return 1, 0, 5, 4. As a general rule of thumb in python: Try to avoid loops at all costs. From my own experience that statement is true in 99 out of 100 cases