CODEWITHSUNDEEP
pythonimage-processingscikit-imagesuperpixels
Steve Emman
Steve Emman

Reputation: 135

Average colour of slic superpixel

I want to segment an image using slic superpixels and then replace the original colour of a superpixel with the average colour of said superpixel.

import numpy as np
import matplotlib.pyplot as plt
from skimage import io
from skimage.segmentation import slic, mark_boundaries
from skimage.data import astronaut
from skimage.measure import regionprops

img = astronaut()
segments = slic(img, n_segments=512, compactness=10,
            multichannel=True,
            enforce_connectivity=True,
            convert2lab=True)
regions = regionprops(segments, intensity_image=img)

I get the errorValueError: Label and intensity image must have thesame shape. Segments shape is (512,512) and img shape in (512,512,3). What is the correct use of regionprops in my case?

Upvotes: 4

Views: 2011

Answers (2)

Cris Luengo
Cris Luengo

Reputation: 60695

According to the documentation, regionprops can only quantify a grey-value image, and won't work for color.

A simple solution would be to measure average intensity in each channel separately, and combine the results:

out = np.empty_like(img)
for ii in range(3):
   regions = regionprops(segments, intensity_image=img[:,:,ii])
   # paint, and write to out[:,:,ii]

Using DIPlib this can be done quite simply (disclaimer: I'm an author):

import diplib as dip

segments = segments.astype('uint32')  # 64-bit types not accepted by DIPlib
msr = dip.MeasurementTool.Measure(segments, img, ['Mean'])
out = dip.ObjectToMeasurement(segments, msr['Mean'])
out.Show()

output of code above

Upvotes: 1

Steve Emman
Steve Emman

Reputation: 135

I followed the first recommendation of the accepted answer. Working version of my code :

import matplotlib.pyplot as plt
from skimage.segmentation import slic
from skimage.data import astronaut
from skimage.measure import regionprops

def paint_region_with_avg_intensity(rp, mi, channel):
    for i in range(rp.shape[0]):
        img[rp[i][0]][rp[i][1]][channel] = mi

img = astronaut()
segments = slic(img, n_segments=512, compactness=10,
            multichannel=True,
            enforce_connectivity=True,
            convert2lab=True)

for i in range(3):
    regions = regionprops(segments, intensity_image=img[:,:,i])
    for r in regions:
        paint_region_with_avg_intensity(r.coords, int(r.mean_intensity), i)

plt.imshow(img)
plt.show()

enter image description here

Upvotes: 4

Related Questions