Blur in LAB format

Question

How can apply a gaussian blur on an image in (CIE)LAB format? RGB can be blurred with rgb_blurred = skimage.filters.gaussian(lab), but that doesn`t work on LAB (because of its first channel). Is there a way to blur images without converting them back to rgb first, and then converting them back?

Answer 1

From the docs:

The multi-dimensional filter is implemented as a sequence of one-dimensional convolution filters.

Therefore you can apply the filter to Lab images too. Images are stored as numpy array, so if you want to apply the filter to some channel(s) only it's no problem using standard numpy indexing. In fact, blurring the a and b channels has little influence on the visual impression. The effect comes from blurring the L channel:

from skimage import data    
from skimage.filters import gaussian
from skimage.color import rgb2lab, lab2rgb
import matplotlib.pyplot as plt

img = data.astronaut()
lab = rgb2lab(img)
blurred = gaussian(lab, 5)
lab[:,:,0] = gaussian(lab[:,:,0], 5, preserve_range=True)

fig, ax = plt.subplots(1,3,figsize=(20,20))
ax[0].imshow(img)
ax[1].imshow(lab2rgb(blurred))
ax[2].imshow(lab2rgb(lab))
ax[0].set_title('Original')
ax[1].set_title('Blurred (entire image)')
ax[2].set_title('Blurred (L channel only)')

Please note that you must set the parameter preserve_range to True when applying the filter to a single channel, otherwise the result will be in the range of 0.0 to 1.0.