shuangyumelody / image-inpainting

Image Inpainting

Implementation of exemplar-based image inpainting algorithm by Criminisi et al.

Requirements:

Python 2.7.9 or greater

Cython 0.22 or greater

NumPy for Python 2

SciPy for Python 2

Matplotlib for Python 2

wxPython 3.0.0 or higher

Instructions

Run the program to open the GUI.

Enter the patch size. By default, it is 9. The patch size must be odd.

Select option to apply Gaussian filtering prior to computing the image gradients and choose sigma value.

Load the image and its mask. Press the Inpaint button to run the algorithm.

Some feasible results

Original image

Mask

Inpainted images

Patch size 9, Gaussian smoothed with sigma=2

Original image

Mask

Inpainted image

Patch size 15, Gaussian smoothed with sigma=1

Original image

Mask

Inpainted image

Patch size 9, Gaussian smooth with sigma=1

Not so good results

Original image

Mask

Inpainted images

Patch size 9, Gaussian smoothed with sigma=1.625

Original image

Mask

Inpainted image

Patch size 19, Gaussian smoothed with sigma=0.5

Patch size 37, no Gaussian smoothing

Original image

Mask

Inpainted image

Patch size 9, Gaussian smoothed with sigma=1.55

Patch size 9, Gaussian smoothed with sigma=1.625