Search code examples
pythonopencvedge-detectioncanny-operator

Is there a way to eliminate the disconnected elements of an image?

I'm working on surface roughness of an image, and I used the edge detection function from opencv, but it also detected the pores from the image. Is there a way to get rid of the pores?

picture of original image and the edges detected

I tried adjusting the threshold values but that only got rid of the edges on the side and made the pores more pronounced. I´m pretty new to opencv and this type of problem, so most of this code is just based on examples I saw online.

import numpy as np
import cv2 as cv
from matplotlib import pyplot as plt
from scipy import ndimage

# need to add the scale to it
img = cv.imread('testim5.png', cv.IMREAD_GRAYSCALE)
blurred = ndimage.filters.gaussian_filter(img, 2, mode='nearest')
edges = cv.Canny(blurred,100,200) # adjusting thresholds will only eliminate the outside edges, not the pores

[testim5.png] https://i.sstatic.net/Vs6K8.png

Solution

  • First, you could use morphological operations to ensure that the interesting edges are connected. You can read more about morphological operations in the web of OpenCV: https://docs.opencv.org/4.x/d9/d61/tutorial_py_morphological_ops.html.

    The code would be:

    import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt
from scipy import ndimage

# Open the image
img = cv.imread('testim5.png', cv.IMREAD_GRAYSCALE)

# Note that `scipy.ndimage.filters.gaussian_filter` is deprecated, use `scipy.ndimage.gaussian_filter` instead
blurred = ndimage.gaussian_filter(img, 2, mode='nearest')
edges = cv.Canny(blurred,100,175)

# Apply morphological operation (closing) to the EDGES so they are connected
kernel = np.ones((5,3),np.uint8)
edges_closing = cv.morphologyEx(edges, cv.MORPH_CLOSE, kernel)

    The result of this is that the vertical edges aren't so disjoint, as you can see in the following image.

    Figure 1

    The second step is to remove the pores. It can be seen as a problem of removing small connected objects and the solution of @Soltius in How to remove small connected objects using OpenCV works well.

    The code in your case would be:

    # Remove smallest connected components
# To understand the following lines, look the answer of @Soltius in https://stackoverflow.com/questions/42798659/how-to-remove-small-connected-objects-using-opencv
nb_blobs, im_with_separated_blobs, stats, _ = cv.connectedComponentsWithStats(edges_closing)
sizes = stats[:, -1]
sizes = sizes[1:]
nb_blobs -= 1

# Minimum size of particles we want to keep (number of pixels).
# Here, it's a fixed value, but you can set it as you want.
min_size = 80  

final_edges = np.zeros_like(im_with_separated_blobs)
for blob in range(nb_blobs):
    if sizes[blob] >= min_size:
        final_edges[im_with_separated_blobs == blob + 1] = 255

fig, axs = plt.subplots(ncols=3)
axs[0].imshow(img, cmap='gray')
axs[1].imshow(edges, cmap='gray')
axs[2].imshow(final_edges, cmap='gray')
axs[0].set_title('Original image')
axs[1].set_title('Original edges')
axs[2].set_title('New edges')
fig.tight_layout()
plt.show()

    The final edges are:

    Figure 2