如何在python中自动检测圆圈

Question

I've been trying to count coins in an images with different backgrounds(not necessarily solid). this is a code I found here in an answer but the problem is I don't want to have to change parameters every image. Is there a way to do this?

def CountCoins_V2(img):
    image_ori = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
    lower_bound = np.array([0,0,10])
    upper_bound = np.array([255,255,195])
    image = img
    mask = cv2.inRange(img, lower_bound, upper_bound)
    #mask = cv2.adaptiveThreshold(image_ori,255,cv2.ADAPTIVE_THRESH_MEAN_C,\cv2.THRESH_BINARY_INV,33,2)
    kernel = np.ones((3, 3), np.uint8)
    mask = cv2.erode(mask, kernel, iterations=6)
    mask = cv2.dilate(mask, kernel, iterations=3)
    closing = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel)
    contours = cv2.findContours(mask.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[1]
    contours.sort(key=lambda x:cv2.boundingRect(x)[0])
    array = []
    ii = 1
    for c in contours:
        (x,y),r = cv2.minEnclosingCircle(c)
        center = (int(x),int(y))
        r = int(r)
        if r >= 6 and r<=10:
            cv2.circle(image,center,r,(0,255,0),2)
            array.append(center)
    show_images([image_ori,mask])
    return len(contours)

Answer 1

There are 2 parameters in the code:

• The first is lower_bound = np.array([0,0,10]) and upper_bound = np.array([255,255,195]) which threshold the image to get only the colors in the range between lower_bound and upper_bound . But in this code, the range is too broad [0,0,10] to [255,255,195] , so it makes no sense. You should limit this range closer to the color of the circles you want to detect. You can read more about color spaces and threshold samples here .
• The second is r in (x,y),r = cv2.minEnclosingCircle(c) , which is the radius of the circle found in the image. It is an important parameter and you need to do experiments to find the best range for all images. My advice is to use ratio between the radius and the width or height of the image, so it will be independent of the resolution of the image.