Чтобы обойти эту проблему, можно использовать следующую процедуру:
- При заданном изображении и маске (значения от 0 до 1)
- Размытие полного входного изображения и маски
- Вес исходного изображения с размытой маской
- Вес размытого изображения с инвертированной размытой маской
- Добавление взвешенных изображений
Нижепример кода с использованием scipy:
import numpy as np
import matplotlib.pyplot as plt
from scipy import misc
import scipy.ndimage
def gaussian_blur(sharp_image, sigma):
# Filter channels individually to avoid gray scale images
blurred_image_r = scipy.ndimage.filters.gaussian_filter(sharp_image[:, :, 0], sigma=sigma)
blurred_image_g = scipy.ndimage.filters.gaussian_filter(sharp_image[:, :, 1], sigma=sigma)
blurred_image_b = scipy.ndimage.filters.gaussian_filter(sharp_image[:, :, 2], sigma=sigma)
blurred_image = np.dstack((blurred_image_r, blurred_image_g, blurred_image_b))
return blurred_image
def uniform_blur(sharp_image, uniform_filter_size):
# The multidimensional filter is required to avoid gray scale images
multidim_filter_size = (uniform_filter_size, uniform_filter_size, 1)
blurred_image = scipy.ndimage.filters.uniform_filter(sharp_image, size=multidim_filter_size)
return blurred_image
def blur_image_locally(sharp_image, mask, use_gaussian_blur, gaussian_sigma, uniform_filter_size):
one_values_f32 = np.full(sharp_image.shape, fill_value=1.0, dtype=np.float32)
sharp_image_f32 = sharp_image.astype(dtype=np.float32)
sharp_mask_f32 = mask.astype(dtype=np.float32)
if use_gaussian_blur:
blurred_image_f32 = gaussian_blur(sharp_image_f32, sigma=gaussian_sigma)
blurred_mask_f32 = gaussian_blur(sharp_mask_f32, sigma=gaussian_sigma)
else:
blurred_image_f32 = uniform_blur(sharp_image_f32, uniform_filter_size)
blurred_mask_f32 = uniform_blur(sharp_mask_f32, uniform_filter_size)
blurred_mask_inverted_f32 = one_values_f32 - blurred_mask_f32
weighted_sharp_image = np.multiply(sharp_image_f32, blurred_mask_f32)
weighted_blurred_image = np.multiply(blurred_image_f32, blurred_mask_inverted_f32)
locally_blurred_image_f32 = weighted_sharp_image + weighted_blurred_image
locally_blurred_image = locally_blurred_image_f32.astype(dtype=np.uint8)
return locally_blurred_image
if __name__ == '__main__':
sharp_image = misc.face()
height, width, channels = sharp_image.shape
sharp_mask = np.full((height, width, channels), fill_value=1)
sharp_mask[int(height / 4): int(3 * height / 4), int(width / 4): int(3 * width / 4), :] = 0
result = blur_image_locally(
sharp_image,
sharp_mask,
use_gaussian_blur=True,
gaussian_sigma=31,
uniform_filter_size=201)
plt.imshow(result)
plt.show()
Результат: