Skip to main navigation Skip to search Skip to main content

Cross-Modal Generative Adversarial Networks for IR-Guided Mosaicked Image Restoration

  • Beihang University
  • Beijing Institute of Remote Sensing Information
  • Key Laboratory of Precision Opto-Mechatronics Technology (Ministry of Education)

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

Abstract

Image mosaicking is widely used for privacy protection but severely degrades visual information in affected regions. This research addresses the challenge of restoring these mosaicked areas in visual-spectrum images. We propose a novel approach that leverages co-registered infrared (IR) imagery, which often remains unaffected by such obfuscation, as a guiding modality. Our method employs a Generative Adversarial Network (GAN) architecture specifically designed to learn the mapping between IR features and corresponding visual content. The GAN is trained to inpaint the mosaicked visual patches by generating realistic details conditioned on the IR image data. Experimental evaluations show the proposed method effectively reconstructs plausible and visually coherent content within the mosaicked regions, significantly enhancing image interpretability. This work demonstrates the potential of cross-modal GANs for robust image restoration in challenging scenarios.

Original languageEnglish
Title of host publicationImage and Graphics Technologies and Applications - 20th Chinese Conference, IGTA 2025, Revised Selected Papers
EditorsYongtian Wang, Yi Chen
PublisherSpringer Science and Business Media Deutschland GmbH
Pages168-182
Number of pages15
ISBN (Print)9789819549658
DOIs
StatePublished - 2026
Event20th Chinese Conference on Image and Graphics Technologies and Applications, IGTA 2025 - Beijing, China
Duration: 9 Aug 202510 Aug 2025

Publication series

NameCommunications in Computer and Information Science
Volume2800 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference20th Chinese Conference on Image and Graphics Technologies and Applications, IGTA 2025
Country/TerritoryChina
CityBeijing
Period9/08/2510/08/25

Keywords

  • Cross-Modal Learning
  • Generative Adversarial Networks
  • Image Demosaicking
  • Image Restoration

Fingerprint

Dive into the research topics of 'Cross-Modal Generative Adversarial Networks for IR-Guided Mosaicked Image Restoration'. Together they form a unique fingerprint.

Cite this