Downsampling without smoothing

Maybe you're looking for resize().

# Python code:
import cv2
large_img = cv2.imread('our_large_image.jpg')
small_to_large_image_size_ratio = 0.2
small_img = cv2.resize(large_img, # original image
                       (0,0), # set fx and fy, not the final size
                       fx=small_to_large_image_size_ratio, 
                       fy=small_to_large_image_size_ratio, 
                       interpolation=cv2.INTER_NEAREST)

Instead of interpolation=cv2.INTER_NEAREST you can use any of these interpolation methods.