Untitled

def sift(image1_path, image2_path, threshold=0.25, ransac_threshold=3.0):
    # Load the images
    if isinstance(image1_path, str):
        img1 = cv2.imread(image1_path, cv2.IMREAD_GRAYSCALE)
    if isinstance(image2_path, str):
        img2 = cv2.imread(image2_path, cv2.IMREAD_GRAYSCALE)
        
    if isinstance(image1_path, np.ndarray):
        img1 = cv2.cvtColor(image1_path.astype('uint8'), cv2.COLOR_RGB2GRAY)
    if isinstance(image2_path, np.ndarray):
        img2 = cv2.cvtColor(image2_path.astype('uint8'), cv2.COLOR_RGB2GRAY)

    # Initialize SIFT detector
    sift = cv2.SIFT_create()

    # Detect and compute keypoints and descriptors
    keypoints1, descriptors1 = sift.detectAndCompute(img1, None)
    keypoints2, descriptors2 = sift.detectAndCompute(img2, None)

    # Initialize a brute-force matcher
    bf = cv2.BFMatcher()

    # Match descriptors
    matches = bf.knnMatch(descriptors1, descriptors2, k=2)

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

    # Convert keypoints to points
    src_pts = [keypoints1[m.queryIdx].pt for m in good_matches]
    dst_pts = [keypoints2[m.trainIdx].pt for m in good_matches]

    # Perform RANSAC for robustness
    if len(src_pts) >= 4:
        src_pts = np.float32(src_pts).reshape(-1, 1, 2)
        dst_pts = np.float32(dst_pts).reshape(-1, 1, 2)

        M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, ransac_threshold)
        mask_matches = mask.ravel().tolist()
        num_matches = sum(mask_matches)
    else:
        num_matches = 0
        
    return num_matches