Let an image and a gap inside it where we know its position, is it possible to erase the gap?
In others words, is it possible from the surrounding information to stretch it inside the gap, and to restore the image as well as possible?. How to strech the surrounding information inside the gap?
We propose : "A Variational Model for Filling-In"(1.2 Mb, PDF)
Some results that we are getting:
Removing gaps on images: (click on the images to zoom)
| Lenna Image with 17 holes |
Reconstruction
Result
|
![]() |
![]() |
Watching our program running:
Superimposed Text Removal. (Every character is considered as a gap)
|
Image
with text to remove
|
Reconstruction
Result
|
![]() |
![]() |
|
Original
Image
|
Object
to remove
|
Our
Result
|
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
Ellipse with a gap(red square). Result of the inpainted gap.
|
Image
with gap to remove
|
Reconstruction
Result
|
![]() |
![]() |
Up, a sequence of numbers from 0 to 9. Middle, the original image with a red hole on digit 6. Bottom, the result inpainting the gap smoothly extending the surrounding information into the hole in the isophote direrections (It's not the solution that we have in mind, but it's the logical solution from the boundary). To restore gaps from surrounding information is enough?. Or, do we need to incorparate knowlegde from Artificial Intelligence ?
![]() |
![]() |
![]() |
An interesting link: http://www.ece.umn.edu/users/marcelo/restoration.html