Resize image with padding using CV2

resize_with_pad_cv2.py

import cv2

from typing import Tuple

def resize_with_pad(image: np.array, 
                    new_shape: Tuple[int, int], 
                    padding_color: Tuple[int] = (255, 255, 255)) -> np.array:
    """Maintains aspect ratio and resizes with padding.
    Params:
        image: Image to be resized.
        new_shape: Expected (width, height) of new image.
        padding_color: Tuple in BGR of padding color
    Returns:
        image: Resized image with padding
    """
    original_shape = (image.shape[1], image.shape[0])
    ratio = float(max(new_shape))/max(original_shape)
    new_size = tuple([int(x*ratio) for x in original_shape])
    image = cv2.resize(image, new_size)
    delta_w = new_shape[0] - new_size[0]
    delta_h = new_shape[1] - new_size[1]
    top, bottom = delta_h//2, delta_h-(delta_h//2)
    left, right = delta_w//2, delta_w-(delta_w//2)
    image = cv2.copyMakeBorder(image, top, bottom, left, right, cv2.BORDER_CONSTANT, value=padding_color)
    return image
  
  if __name__ == "__main__":
    image = cv2.imread("/path/to/image")
    image = resize_with_pad(image, (256, 256))
    
    cv2.imshow("Padded image", image)
    cv2.waitKey()

Thanks man, really helped me today

Thanks man, really helped me today

No Problem!

i like to suggest an optimization to work with any image size, the original code have a problem when the min dimension leng of the image after resize its bigger than the objective size in that dimension, only add this lines:

` Suggestion

if new_size[0] > new_shape[0] or new_size[1] > new_shape[1]:
    ratio = float(min(new_shape)) / min(original_shape)
    new_size = tuple([int(x * ratio) for x in original_shape])

`
that adjusts the ratio to resize the image in the range of objective size

the final function is this:

` Final Fnction

  def resize_with_pad(image: np.array, new_shape: Tuple[int, int], padding_color: Tuple[int] = (255, 255, 255)) -> np.array:
      original_shape = (image.shape[1], image.shape[0])
      ratio = float(max(new_shape))/max(original_shape)
      new_size = tuple([int(x*ratio) for x in original_shape])
  
      if new_size[0] > new_shape[0] or new_size[1] > new_shape[1]:
          ratio = float(min(new_shape)) / min(original_shape)
          new_size = tuple([int(x * ratio) for x in original_shape])
  
      image = cv2.resize(image, new_size)
      delta_w = new_shape[0] - new_size[0]
      delta_h = new_shape[1] - new_size[1]
      top, bottom = delta_h//2, delta_h-(delta_h//2)
      left, right = delta_w//2, delta_w-(delta_w//2)
  
      image = cv2.copyMakeBorder(image, top, bottom, left, right, cv2.BORDER_CONSTANT,None,value=padding_color)
      return image

`

In case of downsizing, delta_w or delta_h will become negative. The following is a fix:

    delta_w = new_shape[0] - new_size[0] if new_shape[0] > new_size[0] else 0
    delta_h = new_shape[1] - new_size[1] if new_shape[1] > new_size[1] else 0

i like to suggest an optimization to work with any image size, the original code have a problem when the min dimension leng of the image after resize its bigger than the objective size in that dimension, only add this lines:

` Suggestion

if new_size[0] > new_shape[0] or new_size[1] > new_shape[1]:
    ratio = float(min(new_shape)) / min(original_shape)
    new_size = tuple([int(x * ratio) for x in original_shape])

` that adjusts the ratio to resize the image in the range of objective size

the final function is this:

` Final Fnction

  def resize_with_pad(image: np.array, new_shape: Tuple[int, int], padding_color: Tuple[int] = (255, 255, 255)) -> np.array:
      original_shape = (image.shape[1], image.shape[0])
      ratio = float(max(new_shape))/max(original_shape)
      new_size = tuple([int(x*ratio) for x in original_shape])
  
      if new_size[0] > new_shape[0] or new_size[1] > new_shape[1]:
          ratio = float(min(new_shape)) / min(original_shape)
          new_size = tuple([int(x * ratio) for x in original_shape])
  
      image = cv2.resize(image, new_size)
      delta_w = new_shape[0] - new_size[0]
      delta_h = new_shape[1] - new_size[1]
      top, bottom = delta_h//2, delta_h-(delta_h//2)
      left, right = delta_w//2, delta_w-(delta_w//2)
  
      image = cv2.copyMakeBorder(image, top, bottom, left, right, cv2.BORDER_CONSTANT,None,value=padding_color)
      return image

`

Thanks! I will test it out and add it to the original.

In case of downsizing, delta_w or delta_h will become negative. The following is a fix:

    delta_w = new_shape[0] - new_size[0] if new_shape[0] > new_size[0] else 0
    delta_h = new_shape[1] - new_size[1] if new_shape[1] > new_size[1] else 0

Will take a look...

Thank you @IdeaKing !