CODEWITHSUNDEEP
pythoncomputer-visioncluster-analysisimage-segmentationscikit-image
maracuja
maracuja

Reputation: 447

Segmentation boundaries generated using Simple Linear Iterative Clustering in skimage are not well defined?

I am using skimage slic clustering algorithm to segment a biomedical image (whole slide image). When I plot the image with the segment boundaries I find that the boundaries are not well defined. Below is the my code and the corresponding image. When I use a even higher resolution image I still have the same problem. Is this because the clustering algorithm can't find well defined segments? Is there a way of obtaining a well defined boundary?

from skimage.segmentation import slic
from skimage.segmentation import mark_boundaries
import matplotlib.pyplot as plt
import cv2

image = cv2.imread('testPatch2.png')
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

segments = slic(image, n_segments=500, sigma=5, enforce_connectivity=True, convert2lab=True)

plt.figure(figsize=(10,10))
plt.imshow(mark_boundaries(image, segments, mode='thick'))
plt.show()

enter image description here

Upvotes: 2

Views: 571

Answers (1)

Juan
Juan

Reputation: 5738

This is not an issue with the segmentation, but with the display. Matplotlib will by default crudely downsample the output image. See this question, those linked therein, and the answers, for more details. You can fix this in different ways:

  1. Set a higher DPI (dots per inch) by passing in dpi=300 (for example) to the plt.figure() call.
  2. Use skimage.io.imsave to save the image returned by mark_boundaries, then open it with a standard image viewer.

Upvotes: 2

Related Questions