Circle detection with OpenCV

Question

How can I improve the performance of the following circle-detection code

from matplotlib.pyplot import imshow, scatter, show
import cv2

image = cv2.imread(\         


        

Answer 1

@Carlos, I'm not really a big fan of Hough Circles in situations like the one you've described. To be honest, I find this algorithm very unintuitive. What I would recommend in your case is using findContour() function and then calculating mass centers. Thus said, I tuned the Hough's parameters a bit to get reasonable results. I also used a different method for preprocessing before Canny, since I don't see how that thresholding would work in any other case than that particular one.

Hough method:

Finding mass centers:

And the code:

from matplotlib.pyplot import imshow, scatter, show, savefig
import cv2

image = cv2.imread('circles.png', 0)
#_, image = cv2.threshold(image, 254, 255, cv2.THRESH_BINARY)
image = cv2.GaussianBlur(image.copy(), (27, 27), 0)
image = cv2.Canny(image, 0, 130)
cv2.imshow("canny", image)
cv2.waitKey(0)
imshow(image, cmap='gray')

circles = cv2.HoughCircles(image, cv2.HOUGH_GRADIENT, 22, minDist=1, maxRadius=50)
x = circles[0, :, 0]
y = circles[0, :, 1]

scatter(x, y)
show()
savefig('result1.png')
cv2.waitKey(0)

_, cnts, _ = cv2.findContours(image.copy(), cv2.RETR_EXTERNAL,
    cv2.CHAIN_APPROX_SIMPLE)

# loop over the contours
for c in cnts:
    # compute the center of the contour
    M = cv2.moments(c)
    cX = int(M["m10"] / M["m00"])
    cY = int(M["m01"] / M["m00"])

    #draw the contour and center of the shape on the image
    cv2.drawContours(image, [c], -1, (125, 125, 125), 2)
    cv2.circle(image, (cX, cY), 3, (255, 255, 255), -1)

cv2.imshow("Image", image)
cv2.imwrite("result2.png", image)
cv2.waitKey(0)

Both methods require some more fine tuning but I hope that gives you something more to work with.

Sources: this answer and pyimagesearch.

Answer 2

While it is possible to fine-tune Hough Circles for a given image, the optimal parameters from image to image may vary alot. Hence, it takes quite some effort to make the circle detection robust using Hough Circles, though its doable!

Instead I would suggest to use more modern approaches like:

• Arc Adjacency Matrix-Based Fast Ellipse Detection (https://ieeexplore.ieee.org/document/8972900). Code available: https://github.com/Li-Zhaoxi/AAMED
• Circle Detection by Arc-support Line Segments ( https://ieeexplore.ieee.org/document/8296246 ). Code available: https://github.com/AlanLuSun/Circle-detection