Image inpainting by low-rank tensor decomposition and multidirectional search

Liu, Xuya; Hao, Caiyan; Su, Zezhao; Qi, Zerong; Fu, Shujun; Li, Yuliang; Han, Hongbin

doi:10.1117/1.jei.30.5.053010

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Image inpainting is a restoration process that estimates the content of missing regions within images and videos and has sparked special interest in the area of computer vision (Elharrouss et al, 2020;Jam et al, 2020). Conventional inpainting algorithms use well known mathematical and statistical approximation methods such as biharmonic functions (Damelin & Hoang, 2018), Cahn-Hilliard equations (Bertozzi et al, 2006), anisotropic diffusion by partial differential equation modeling for propagating boundary data (Bertalmio et al, 2000), low-rank tensor completion (Li et al, 2020;Liu et al, 2021) and Gauss-Markov random field priors (Satapathy & Sahay, 2021;Efros & Leung, 1999). Alternatively, machine learning (ML) approaches have potential for inpainting tasks.…”

Section: Introductionmentioning

confidence: 99%

A comparison of deep-learning-based inpainting techniques for experimental X-ray scattering

et al. 2022

View full text Add to dashboard Cite

The implementation is proposed of image inpainting techniques for the reconstruction of gaps in experimental X-ray scattering data. The proposed methods use deep learning neural network architectures, such as convolutional autoencoders, tunable U-Nets, partial convolution neural networks and mixed-scale dense networks, to reconstruct the missing information in experimental scattering images. In particular, the recovered pixel intensities are evaluated against their corresponding ground-truth values using the mean absolute error and the correlation coefficient metrics. The results demonstrate that the proposed methods achieve better performance than traditional inpainting algorithms such as biharmonic functions. Overall, tunable U-Net and mixed-scale dense network architectures achieved the best reconstruction performance among all the tested algorithms, with correlation coefficient scores greater than 0.9980.

show abstract