Superpixels

This notebook expands on the idea of splitting the image in superpixels and afterwards classifying each superpixel depending on its texture.



In [1]:

    
%pylab inline









    



Populating the interactive namespace from numpy and matplotlib



In [2]:

    
from skimage.segmentation import slic
from skimage.util import img_as_float
from skimage.segmentation import mark_boundaries
from skimage import io

Image

First, I'm just loading a random image (this is a concept test, so no cancer on this one)



In [3]:

    
image = img_as_float(io.imread('IMG_3721.jpg'))



In [4]:

    
figure(figsize=(10,10))
imshow(image)
axis('off');

Second

Now, I'm splitting the image in a maximum of 200 super pixels



In [5]:

    
numSegments = 200
segments = slic(image, n_segments = numSegments, sigma = 5)
figure(figsize=(10,10))
imshow(mark_boundaries(image, segments))
axis('off')
;









    Out[5]:





''

Basically, each superpixel is a region of pixels which are similar. The segments are a matrix of pixel labels, where segments[i,j] = N means that pixel i,j belongs in superpixel N.



In [6]:

    
imshow(segments)
axis('off')
colorbar();

Finally

So we can cut out superpixels:



In [11]:

    
mask = segments==40
masked = image.copy()
for i in range(3):
    masked[:,:,i] = masked[:,:,i] * mask
x,y = np.where(mask)
imshow(masked[x.min():x.max(), y.min():y.max(), :])
axis('off');



In [12]:

    
mask = segments==60
masked = image.copy()
for i in range(3):
    masked[:,:,i] = masked[:,:,i] * mask
x,y = np.where(mask)
imshow(masked[x.min():x.max(), y.min():y.max(), :])
axis('off');



In [13]:

    
mask = segments==100
masked = image.copy()
for i in range(3):
    masked[:,:,i] = masked[:,:,i] * mask
x,y = np.where(mask)
imshow(masked[x.min():x.max(), y.min():y.max(), :])
axis('off');

We could now classify each super pixel as Zone 1, zone 2, zone3.