[Python] Finding contours in noisy images

I try to find the inner boundaries within this 8-bit grayscale image:

img = cv2.imread("C:/users/XXX/test.png", -1)




I've been playing around with various morphological operations but I can't seem to find any that work well. The most promising approach so far:

After (simple) binarization I receive the following:

img = (img > 0) * 1
img = img.astype('uint8')
plt.imshow(img, cmap = "gray", vmin = 0, vmax = 1)
plt.show()
plt.close()




Or, using mild blurring in advance:

img = cv2.medianBlur(img, 3)
img = (img > 0) * 1
img = img.astype('uint8')
plt.imshow(img, cmap = "gray", vmin = 0, vmax = 1)
plt.show()
plt.close()




Applying closing with an asymmetric kernel I then tried to further enhance the more-or-less horizontal lines:

kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (15,3))
imgClose = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
plt.imshow(imgClose, cmap = "gray", vmin = 0, vmax = 1)
plt.show()
plt.close()




What I want to achieve (red line drawn by hand):



Any idea how to get there?

Continue reading...