I work with the interfaces of various programs on my computer, I find elements that I can interact with, for example buttons. At some point, I have accumulated a large number of buttons, and I need an optimal algorithm for finding them in this image.

I write in python, I tried cv2.matchTemplate – it just matches the template, and it seemed to me that this is not the optimal method, because I just have to go through the entire database, then I tried the methods of searching for key points: surf, sift, orb, their problem is that they do not always find key points for small buttons (and there are many of them in MS Word). I also tried to look for the contours of the images, and to match them, but I also could not come up with an optimal method.

The code example below cannot find the key points for the button.

enter image description here

enter image description here

img1 = cv2.imread( 'InputPhoto/6-2.jpg' , cv2.IMREAD_GRAYSCALE) 
img2 = cv2.imread( "InputPhoto/6-1.jpg" , cv2.IMREAD_GRAYSCALE) 

sift = cv2.SIFT_create() 
keypoints1, descriptors1 = sift.detectAndCompute(img1, None ) 
keypoints2, descriptors2 = sift.detectAndCompute(img2, None ) 

bf = cv2.BFMatcher() 
matches = bf.knnMatch(descriptors1, descriptors2, k= 2) 

threshold = 0.55

good_matches = [] 
for m, n in matches: 
    if m.distance < threshold * n.distance: 
        good_matches.append(m) 

img_matches = cv2.drawMatches(img1, keypoints1, img2, keypoints2, good_matches, None , flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS) 

plt.figure(figsize = ( 20 , 10 )) 
plt.imshow(img_matches)