I have array of floating points Nx2 which is representing the reprojected 3D -> 2D coordinates in a 2D image. I need to find closest (integer) pixel. The naive solution is pretty simple:

import numpy as np

# Original array of floating-point row and column values
points = np.array([
    [327.47910325, 3928.36321963],
    [1439.79734793, 3987.02652005],
    [304.02698845, 2844.82490694],
    [230.43053757, 4090.70452501]
])

def find_nearest_integer_point(point):
    # Extract the row and column values
    row, col = point
    
    # Four possible nearest integer points
    candidates = np.array([
        [np.floor(row), np.floor(col)],
        [np.floor(row), np.ceil(col)],
        [np.ceil(row), np.floor(col)],
        [np.ceil(row), np.ceil(col)]
    ])
    
    # Calculate the Euclidean distances to the candidates
    distances = np.linalg.norm(candidates - point, axis=1)
    
    # Select the candidate with the minimum distance
    nearest_point = candidates[np.argmin(distances)]
    
    return nearest_point.astype(int)

# Apply the function to each point
rounded_points = np.array([find_nearest_integer_point(point) for point in points])

print(rounded_points)

But this for sure will be extremely inefficient once working with big amounts of data (2D videos instead of a single image).
Anyone could advise me on a some speed-up for the given method? Maybe something already implemented in some publicly available library? Or maybe some data structure I should take into account?