Large Region Inpainting by Re-weighted Regularized Methods, Inverse Problems and Imaging

Yiting Chen, Jia Li, Qingyun Yu

Abstract:

In the development of imaging science and image processing request
in our daily life, inpainting large regions always plays an important role.
However, the existing local regularized models and some patch manifold based
non-local models are often not eective in restoring the features and patterns in
the large missing regions. In this paper, we will apply a strategy of inpainting
from outside to inside and propose a re-weighted matching algorithm by closest
patch (RWCP), contributing to further enhancing the features in the missing
large regions. Additionally, we propose another re-weighted matching algorithm
by distance-based weighted average (RWWA), leading to a result with
higher PSNR value in some cases. Numerical simulations will demonstrate
that for large region inpainting, the proposed method is more applicable than
most canonical methods. Moreover, combined with image denoising methods,
the proposed model is also applicable for noisy image restoration with large
missing regions.