Multi‐stage image inpainting using improved partial convolutions

Abstract In recent years, deep learning models have dramatically influenced image inpainting. However, many existing studies still suffer from over‐smoothed or blurred textures when missing regions are large or contain rich visual details. To restore textures at a fine‐grained level, a multi‐stage i...

Full description

Saved in:
Bibliographic Details
Main Authors: Cheng Li, Dan Xu, Hao Zhang
Format: Article
Language:English
Published: Wiley 2024-10-01
Series:IET Image Processing
Subjects:
Online Access:https://doi.org/10.1049/ipr2.13178
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850200371182436352
author Cheng Li
Dan Xu
Hao Zhang
author_facet Cheng Li
Dan Xu
Hao Zhang
author_sort Cheng Li
collection DOAJ
description Abstract In recent years, deep learning models have dramatically influenced image inpainting. However, many existing studies still suffer from over‐smoothed or blurred textures when missing regions are large or contain rich visual details. To restore textures at a fine‐grained level, a multi‐stage inpainting approach is proposed, which applies a series of partial inpainting modules as well as a progressive inpainting module to inpaint missing areas from their boundaries to the centre successively. Some improvements are made on the partial convolutions to reduce artifacts like blurriness, which require a convolution kernel to contain known pixels more than a certain proportion. Towards photorealistic inpainting results, the intermediate outputs from each stage are used to compute the loss. Finally, to facilitate the training process, a multi‐step training is designed that progressively adds inpainting modules to optimize the model. Experiments show that this method outperforms the current excellent techniques on the publicly available datasets: CelebA, Places2 and Paris StreetView.
format Article
id doaj-art-0da85b4d2d2c438f9576de894c9e42ec
institution OA Journals
issn 1751-9659
1751-9667
language English
publishDate 2024-10-01
publisher Wiley
record_format Article
series IET Image Processing
spelling doaj-art-0da85b4d2d2c438f9576de894c9e42ec2025-08-20T02:12:20ZengWileyIET Image Processing1751-96591751-96672024-10-0118123343335510.1049/ipr2.13178Multi‐stage image inpainting using improved partial convolutionsCheng Li0Dan Xu1Hao Zhang2School of Information Science and Engineering Yunnan University Kunming ChinaSchool of Information Science and Engineering Yunnan University Kunming ChinaSchool of Information Science and Engineering Yunnan University Kunming ChinaAbstract In recent years, deep learning models have dramatically influenced image inpainting. However, many existing studies still suffer from over‐smoothed or blurred textures when missing regions are large or contain rich visual details. To restore textures at a fine‐grained level, a multi‐stage inpainting approach is proposed, which applies a series of partial inpainting modules as well as a progressive inpainting module to inpaint missing areas from their boundaries to the centre successively. Some improvements are made on the partial convolutions to reduce artifacts like blurriness, which require a convolution kernel to contain known pixels more than a certain proportion. Towards photorealistic inpainting results, the intermediate outputs from each stage are used to compute the loss. Finally, to facilitate the training process, a multi‐step training is designed that progressively adds inpainting modules to optimize the model. Experiments show that this method outperforms the current excellent techniques on the publicly available datasets: CelebA, Places2 and Paris StreetView.https://doi.org/10.1049/ipr2.13178image inpaintingimage processingimage reconstructionmulti‐step training processpartial inpainting moduleprogressive inpainting module
spellingShingle Cheng Li
Dan Xu
Hao Zhang
Multi‐stage image inpainting using improved partial convolutions
IET Image Processing
image inpainting
image processing
image reconstruction
multi‐step training process
partial inpainting module
progressive inpainting module
title Multi‐stage image inpainting using improved partial convolutions
title_full Multi‐stage image inpainting using improved partial convolutions
title_fullStr Multi‐stage image inpainting using improved partial convolutions
title_full_unstemmed Multi‐stage image inpainting using improved partial convolutions
title_short Multi‐stage image inpainting using improved partial convolutions
title_sort multi stage image inpainting using improved partial convolutions
topic image inpainting
image processing
image reconstruction
multi‐step training process
partial inpainting module
progressive inpainting module
url https://doi.org/10.1049/ipr2.13178
work_keys_str_mv AT chengli multistageimageinpaintingusingimprovedpartialconvolutions
AT danxu multistageimageinpaintingusingimprovedpartialconvolutions
AT haozhang multistageimageinpaintingusingimprovedpartialconvolutions