System information (version)
- OpenCV => 3.3.0 master
- Operating System / Platform => Ubuntu 16.04 / x86_64
- Compiler => gcc 5.4
Detailed description
GaussianBlur produces darker results than the input image.
Original Image:
>>> cv2.imwrite('gaus.30.cv2.png', cv2.GaussianBlur(im, (151, 151), 30))
$ gm convert gaus.0.png -blur 75x30 gaus.30.gm.png
Images should look the same, but this is obvious, that produced by OpenCV is much darker.
Proof that results produced by OpenCV is wrong
In theory, Gaussian blur shouldn't change color distribution or lightness. So, lets calculate lightness of the images:
# Using Pillow
from PIL import Image
def lightness(im):
hist = im.histogram()
return [
sum(i*v for i, v in enumerate(hist[:256])) / im.width / im.height,
sum(i*v for i, v in enumerate(hist[256:512])) / im.width / im.height,
sum(i*v for i, v in enumerate(hist[512:])) / im.width / im.height
]
# [59.76, 69.08, 40.2]
print(lighnes(Image.open('gaus.0.png')))
# [59.73, 69.18, 39.94]
print(lighnes(Image.open('gaus.30.gm.png')))
# [53.86, 62.26, 36.22]
print(lighnes(Image.open('gaus.30.cv2.png')))
System information (version)
Detailed description
GaussianBlurproduces darker results than the input image.Original Image:
Images should look the same, but this is obvious, that produced by OpenCV is much darker.
Proof that results produced by OpenCV is wrong
In theory, Gaussian blur shouldn't change color distribution or lightness. So, lets calculate lightness of the images: