Examplar-based inpainting based on local geometry

O. Le Meur, J. Gautier and C. Guillemot
[pdf].[pdf poster]

Home Main idea Original figure of the paper Supplementary materials Software Bibtex

Main Idea

We propose a novel inpainting algorithm combining the advantage of PDE-based schemes and examplar-based approaches. The proposed algorithm relies on the use of structure tensors and template matching. The structure tensors are computed in a hierarchic manner whereas the template matching is based on a K-nearest neighbor algorithm. The value K is adaptively set in function of the local texture information. Compared to two state of the art approaches, the proposed method provides more coherent results.
The video below shows how the proposed approach works. The hierarchical option was here disabled for the sake of visibility. Note that the first inpainted pictures have a resolution of 200x200 whereas the latest have a resolution close to 500x400 pixels.

Original figure of the paper

Click on the pictures to look at the original picture.

• Figure 1:

From left-hand side to the right-hand side: direction of the isophotes; coherence norm: black areas correspond to areas for which there is no dominant direction; Filling with the best candidate (K=1); Filling with the best 10 candidates.


• Figure 2:

Comparison of the proposed approach with Criminisi et al.'s, Tschumperlé et al.'s and Patch Match (not presented in the paper) approach.








• Figure 3:

Zoom on the picture of figure 2 (last row) without the hierarchical approach (a) and with it (b).




• Figure 4:

Results of the proposed approach on pictures proposed by Kawai et al.

Supplementary materials

Masks and original pictures:

• Masks and pictures of the paper: [dataset]
• Others: [dataset] From N. Kawai, T. Sato, and N. Yokoya, Image inpainting considering brightness change and spatial locality of textures and its evaluation, in PSIVT2009, 2009, pp. 271–282.

• Isabe : 50% of the pictures was removed. The quality of the reconstrution [ PSNR = 24 dB (25.28dB with hierarchical approach (3 levels))]

• Gorillas :

• Lena picture: 50% of the picture was removed. The quality of the reconstrution [PSNR = 26.8 dB (28.81dB with hierarchical approach (3 levels))]

• Sailing picture: 50% of the picture was removed. The quality of the reconstrution [PSNR = 24 dB (24.44dB with hierarchical approach (3 levels))]

• Women picture: 50% of the picture was removed. The quality of the reconstrution [PSNR = 23.44 dB (24.38dB with hierarchical approach (3 levels))]

Software

Not yet available.

BibTex

@InProceedings{LeMeur_2011,
  author = {O. {Le Meur} and J. Gautier and C. Guillemot},
  title = {Examplar-based inpainting based on local geometry},
  booktitle = {ICIP},
  year = {2011}
}