Image and Video Restoration

Learning-based frameworks to recover images and video from corrupted flash storage. National Research Foundation, Singapore (NRF) · Jan 2021 – Mar 2024

Funding: National Research Foundation, Singapore (NRF) · Jan 2021 – Mar 2024

Flash storage corruption leaves JPEG images and video streams in states that standard decoders cannot handle. This project designed learning-based frameworks to recover corrupted images and videos from flash storage, enabling more reliable data recovery.

Contributions:

  • Designed a two-stage compensation and alignment framework for JPEG image restoration from corrupted bitstreams (CVPR 2023).
  • Contributed to the first public benchmark dataset and recovery method for bitstream-corrupted video (NeurIPS 2023).
  • Proposed a spatial-focal error concealment scheme for corrupted focal-stack video (DCC 2023).
  • Extended byte-level representation to multimedia file fragment classification using visual perspectives (IEEE TMM 2024).

Related publications: CVPR 2023 · NeurIPS 2023 · DCC 2023 · IEEE TMM 2024 · IEEE TMM 2026 · AICAS 2023 · ISCAS 2023