import cv2
def get_best_images(plate_images, num_img_return):
"""
Get the top num_img_return quality images (with the least blur).
Laplacian function returns a value which indicates how blur the image is.
The lower the value, the more blur the image have
"""
# first, pick the image with the largest area because the bigger the image, the bigger the characters on the plate
if len(plate_images) > (num_img_return + 2):
plate_images = sorted(plate_images, key=lambda x : x[0].shape[0]*x[0].shape[1], reverse=True)[:(num_img_return+2)]
# secondly, pick the images with the least blur
if len(plate_images) > num_img_return:
plate_images = sorted(plate_images, key=lambda img : cv2.Laplacian(img[0], cv2.CV_64F).var(), reverse=True)[:num_img_return]
# img[0] because plate_images = [plate image, char on plate]
return plate_images
