Crop the center of the image in Keras ImageDataGenerator or flow_from_directory

Question

I am trying to crop the center of the images in image data generator using keras. I have images of size 192x192 and I want to crop the center of them so that the output batches will be 150x150 or something similar.

Could I do this immediately in Keras ImageDataGenerator? I guess no, since I saw the the target_size argument in datagenerator smashes the images.

I found this link for random cropping: https://jkjung-avt.github.io/keras-image-cropping/

I have already modified the crop as follows:

def my_crop(img, random_crop_size):
  if K.image_data_format() == 'channels_last':
    # Note: image_data_format is 'channel_last'
    assert img.shape[2] == 3
    height, width = img.shape[0], img.shape[1]
    dy, dx = random_crop_size #input desired output size
    start_y = (height-dy)//2
    start_x = (width-dx)//2
    return img[start_y:start_y+dy, start_x:(dx+start_x), :]
  else:
      assert img.shape[0] == 3
      height, width = img.shape[1], img.shape[2]
      dy, dx = random_crop_size  # input desired output size
      start_y = (height - dy) // 2
      start_x = (width - dx) // 2
      return img[:,start_y:start_y + dy, start_x:(dx + start_x)]

def crop_generator(batches, crop_length):
    '''
    Take as input a Keras ImageGen (Iterator) and generate
    crops from the image batches generated by the original iterator
    '''
    while True:
        batch_x, batch_y = next(batches)
       #print('the shape of tensor batch_x is:', batch_x.shape)
        #print('the shape of tensor batch_y is:', batch_y.shape)
        if K.image_data_format() == 'channels_last':
         batch_crops = np.zeros((batch_x.shape[0], crop_length, crop_length, 3))
        else:
         batch_crops = np.zeros((batch_x.shape[0], 3, crop_length, crop_length))
        for i in range(batch_x.shape[0]):
            batch_crops[i] = my_crop(batch_x[i], (crop_length, crop_length))
        yield (batch_crops, batch_y)

This solution seems to me very slow, please is there any other more efficient way? what would you suggest?

Thanks in Advance

Crossgarnet

Answer 1

I tried to solve it in this way:

def crop_generator(batches, crop_length):
  while True:
    batch_x, batch_y = next(batches)
    start_y = (img_height - crop_length) // 2
    start_x = (img_width - crop_length) // 2
    if K.image_data_format() == 'channels_last':
        batch_crops = batch_x[:, start_x:(img_width - start_x), start_y:(img_height - start_y), :]
    else:
        batch_crops = batch_x[:, :, start_x:(img_width - start_x), start_y:(img_height - start_y)]
    yield (batch_crops, batch_y)

still if you have better way please give your suggestions.

Crossgarnet

Answer 2

You can define your custom cropping function as something like this.

from keras.preprocessing.image import load_img
from keras.preprocessing.image import load_img
from keras.preprocessing.image import img_to_array
from keras.preprocessing.image import array_to_img

def crop(image):
    
    start_y = (img_height - crop_length) // 2
    start_y = (img_width - crop_length) // 2
    cropped_image=image[start_x:(img_width - start_x), start_y:(img_height - start_y), :]
    
    return cropped_image

The function can be included in Keras Custom generator like this.

ImageDataGenerator(preprocessing_function=crop) 

And then you can ImageDataGenerator object as per your choice

Kitchenware