I have this image: HI00008918.png
I want to apply a logarithmic function (f(x) = (1/a)*log(x + 1), where a = 0.01) on the image...
So this is the code:
import numpy as np
import matplotlib.pyplot as plt
import skimage.io as io
car = io.imread('HI00008918.png')
# plt.imshow(car, cmap='gray', vmin=0, vmax=255)
a = 0.01
fnLog = lambda x : (1/a)*np.log(x + 1) # logarithmic function
# the original image has white pixels (=0) and black pixels (=255)
carLog = fnLog(car) # Applying the function fnLog
print(car[0][0][-1])
print(carLog[0][0][-1])
print(fnLog(car[0][0][-1]))
The output:
255
-inf
554.5177444479563
Look at one moment it results in -inf and at others it results in the correct value :(
Now I will show the arrays:
carLog =
[[[277.2 277.2 277.2 -inf]
[289. 289. 289. -inf]
[304.5 304.5 304.5 -inf]
...
[423.5 431.8 429. -inf]
[422. 434.5 427.8 -inf]
[437. 450. 440.5 -inf]]
[[434.5 434.5 434.5 -inf]
[433.2 433.2 433.2 -inf]
[430.5 430.5 430.5 -inf]
...
[422. 430.5 427.8 -inf]
[420.2 429. 426.2 -inf]
[433.2 444.2 438.2 -inf]]]
car =
[[[ 15 15 15 255]
[ 17 17 17 255]
[ 20 20 20 255]
...
[148 138 149 255]
[138 125 142 255]
[148 134 151 255]]
[[ 10 10 10 255]
[ 14 14 14 255]
[ 19 19 19 255]
...
It looks like np.log(x + 1) gives -Inf only where x is 255 in an array.
Because the array x is uint8, adding 1 to 255 causes overflow, which wraps the result yielding 0. The log of 0 is -Inf.
You might want to cast the image to a floating-point type before applying the function:
carLog = fnLog(car.astype(np.float32))
When you apply the function to a value extracted from the image, you are working with a Python int, which doesn’t ever overflow.