Image Processing Tutorial Using scikit-image - Image Restoration
Image Restoration:
In this short tutorial, we will employ skimage.restoration.inpaint_biharmonic function to undo the severe damage as much as possible in the following image:
#Import necessary libraries
import skimage.io
from skimage.restoration import inpaint
import matplotlib.pyplot as plt
import numpy as np
from skimage import transform
import skimage.filters#Load damaged image:
damaged_image = skimage.io.imread('damaged_image.jpg')
#Rescale the image
image_rescaled = transform.rescale(damaged_image, 1/4,anti_aliasing=False, multichannel=True )
#Display the image
skimage.io.imshow(image_rescaled)
plt.title('Original Image')
plt.axis('off')(-0.5, 511.5, 717.5, -0.5)
First, we need to denoise the image:
filtered_image = skimage.filters.gaussian(image_rescaled,sigma=2)
skimage.io.imshow(filtered_image)
plt.title('Denoised Image')
plt.axis('off')<ipython-input-53-36013d7c00b5>:1: RuntimeWarning: Images with dimensions (M, N, 3) are interpreted as 2D+RGB by default. Use `multichannel=False` to interpret as 3D image with last dimension of length 3. filtered_image = skimage.filters.gaussian(image_rescaled,sigma=2)
(-0.5, 511.5, 717.5, -0.5)
As we can see, some parts of the image are torn out. Our focus is to restore the white parts of the image as much as possible. For this, we will create a binary mask with the same shape as the image. The mask's pixels that correspond to the torn parts of the image will have a value of 1 (white), the rest will be 0 (black). The mask will then be used by inpaint_biharmonic() to restore the torn parts.
#Create a mask. The image's torn out parts will be
#white in the mask, the rest of the mask will be black.
mask = np.zeros(image_rescaled.shape[:-1])
mask[360:400,40:100]=1
mask[382:405, 330:357]=1
mask[315:435, 355:488]=1
#Display the mask:
skimage.io.imshow(mask)
plt.title('Binary Mask')
plt.axis('off')(-0.5, 511.5, 717.5, -0.5)
#Restore the image:
restored_image = skimage.restoration.inpaint_biharmonic(filtered_image, mask, multichannel=True)
#Plot the restored image along with the original image:
f, ax = plt.subplots(1,2)
ax[0].imshow(image_rescaled)
ax[0].set_title('Original Image')
ax[0].axis('off')
ax[1].imshow(restored_image)
ax[1].set_title('Restored Image')
ax[1].axis('off')(-0.5, 511.5, 717.5, -0.5)
Although, the restored image is still in need of more restoration; compared to the original one, we see a lot of improvements, especially on the right hand side of the image.