Why is Local Binary Pattern affected by scaling?

Question

I found that the local binary pattern in scikit-image is affected by re-scaling the image, but I was not expecting this. Since the LBP just involves greater/less than comparisons between nearby pixels, I thought a linear transformation would not affect things.

import numpy as np
import skimage
from skimage.feature import local_binary_pattern

im = skimage.data.cell()

out_im = local_binary_pattern(im, 16, 2, method='uniform')
out_im = out_im[20:-20, 20:-20] # remove edges
counts,bins = np.histogram(out_im)
print('counts=',counts)
print('bins=',bins)

This gives me

counts= [   896   2743   9555  14928 108466  94041  28682  14703   8728  33458]
bins= [ 0.   1.7  3.4  5.1  6.8  8.5 10.2 11.9 13.6 15.3 17. ]

But if I normalize the image:

im = (im-im.min())/(im.max()-im.min())

Then I get:

counts= [   937   2716   9504  16263 109302  92114  27753  14248   8667  34696]
bins= [ 0.   1.7  3.4  5.1  6.8  8.5 10.2 11.9 13.6 15.3 17. ]

Can someone explain why? The original image has values between 0 and 255.

The comparison is done on fixed positions. When you resize the image, you will compare with different content (the same fixed positions have a different content now) — Micka
– Micka, Commented Aug 7, 2022 at 22:47
I’m not resizing, just mapping the intensity to the range [0,1] — user3433489
– user3433489, Commented Aug 8, 2022 at 12:03
Can you tell min and max values? What happens in the image border, if 0 neighborhood is assumed outside of the image you will have additional "equal" patterns after scaling min to 0 — Micka
– Micka, Commented Aug 8, 2022 at 12:45
I added them in the post. They are 0 and 255. I also removed the edges. — user3433489
– user3433489, Commented Aug 9, 2022 at 15:40

user3433489 · Accepted Answer · 2022-08-09 15:50:51Z

1

I found that there is no difference when there are only 4 points in the kernel, which correspond to up/down/left/right. This suggests that the difference arises due to the floating point interpolation when the points are located at non-integer coordinates. I'm guessing that the floating point calculations occasionally mess up the greater/equal comparisons, leading to slightly different histograms.

answered Aug 9, 2022 at 15:50

user3433489

1,0092 gold badges12 silver badges27 bronze badges

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

2 Comments

Cris Luengo Over a year ago

I think you're right, this is a floating-point rounding error issue. Note that the scaling itself already introduces rounding errors.

user3433489 Over a year ago

There are sometimes even differences when multiplying an integer image by another integer. The image remains integer-type, but if the number of kernel points is not 4, the floating point interpolation sometimes changes things.

Collectives™ on Stack Overflow

Why is Local Binary Pattern affected by scaling?

1 Answer 1

2 Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Related