How to set overflowed subtraction in python to result in a zero?

Question

I am trying to find regions which have features strikingly different from the baseline.

To do that I subtract current image from the base, f and b are grayscale 2d image matrices.

diff = f - b

Some of the oeprations result in overflow and this leads to regions of high pixel value where really they should be set to zero.

How do I specify that the operation diff = f - b should yield 0 for individual pixel value if f[x][y] < b[x][y] ?

Clamping the difference to 0 would make changes in one direction invisible. Perhaps you want `abs(f - b)`? — jasonharper, Apr 07 '18 at 18:49
I do, but I want a function for matrices and I'd rather not write one myself as speed would probably suck. — mega_creamery, Apr 07 '18 at 20:05

Paul Panzer · Accepted Answer · 2018-04-08T01:12:31.773

4

Here is one way of doing it in numpy that doesn't require casting to a larger integer type:

f - b.clip(None, f)

or, equivalently,

f - np.minimum(b, f)

edited Apr 08 '18 at 01:12

answered Apr 08 '18 at 01:05

Paul Panzer

51,835
3
54
99

1

very elegant solution! – Georg W. Apr 29 '20 at 12:07

score 0 · Answer 2 · answered Apr 07 '18 at 22:01

I fixed that by making my own function which compares each pixel before subtracting it to prevent any overflow from occuring.

def custom_sub(i2,i1):      
    x =  len(i1)
    y =  len(i1[0])

    o = deepcopy(i1)

    for ix in range(x):             
            for iy in range(y):
                if i1[ix][iy] > i2[ix][iy]:
                    o[ix][iy] = 0
                else:
                    o[ix][iy] = i2[ix][iy]-i1[ix][iy]
    return o

This is the output, bright regions now only where the spark flash is occurring.

This question is similiar to mine. Subtraction in my case was performed with uint8 type, this can be converted to int16. Resulting picture can then be iterated over and any negative number removed.

Indeed, signed difference is the right approach. Numpy has a clipping function too. `np.clip` or something like that. Would be easier than looping over all pixels. — Cris Luengo, Apr 08 '18 at 00:51

How to set overflowed subtraction in python to result in a zero?

2 Answers2