Research on Image Restoration Algorithms Based on Transformer Prediction Networks

Qingjuan Wang*, Shan Xiao, Yan Zhou, Lin Wang

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In recent years, with the rapid development of computer vision, there have been significant advancements in image restoration techniques. However, many current algorithms still struggle with accurately restoring degraded images with lost information. In this paper, we present a new image restoration algorithm called CodeFormerGAN, which is based on the Transformer model. This algorithm consists of three main components: Codebook, CLT network, and CFT. The Codebook module is responsible for learning self-reconstruction and storing high-quality face image parts. The Transformer in the CLT network models the overall composition of the face using low-quality input. The CFT module transforms image features to achieve a flexible balance between restoration quality and fidelity. Additionally, we introduce a GAN-based image super-resolution enhancement algorithm to enhance the sharpness of the restored image. The results of our experiments demonstrate that our proposed method produces higher-quality restoration results and exhibits better robustness when processing images in complex scenes.

Original languageEnglish
Title of host publication2023 IEEE 4th International Conference on Pattern Recognition and Machine Learning, PRML 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages243-249
Number of pages7
ISBN (Electronic)9798350324303
DOIs
Publication statusPublished - 2023
Event4th IEEE International Conference on Pattern Recognition and Machine Learning, PRML 2023 - Urumqi, China
Duration: 4 Aug 20236 Aug 2023

Publication series

Name2023 IEEE 4th International Conference on Pattern Recognition and Machine Learning, PRML 2023

Conference

Conference4th IEEE International Conference on Pattern Recognition and Machine Learning, PRML 2023
Country/TerritoryChina
CityUrumqi
Period4/08/236/08/23

Keywords

  • Codeformer
  • GAN
  • Image Restoration
  • Transformer

Fingerprint

Dive into the research topics of 'Research on Image Restoration Algorithms Based on Transformer Prediction Networks'. Together they form a unique fingerprint.

Cite this