Home // SIGNAL 2025, The Tenth International Conference on Advances in Signal, Image and Video Processing // View article

FMSTFnet: Feature-Modulation Spatio-Temporal Fusion Network for HDR Video

Authors:
Wei Zhang
Yeyao Chen
Gangyi Jiang

Keywords: High dynamic range video reconstruction, feature modulation, spatio-temporal fusion, transformer block

Abstract:
This paper proposes a novel High Dynamic Range video (HDRv) reconstruction method from Standard Dynamic Range video (SDRv), with a feature modulation spatio-temporal fusion network (FMSTFnet). FMSTFnet has low-frequency and high-frequency parts with a pyramid structure. The low-frequency part mainly includes a Combined Global and Local Feature Modulation module (CGLFM) and a Spatio-Temporal Fusion Module (STFM). CGLFM modulates global and local features of SDR frames to correct the detail deviation caused by brightness differences in different regions and obtain preliminary HDR frames. STFM is designed to enhance the preliminary HDR frames using inter-frame information, and eliminate possible inter-frame artifacts. Finally, an adaptive hybrid module is constructed to fuse the low-frequency HDR frames and gradually extend the processed high-frequency information from low resolution to the higher. The proposed network fully utilizes the inter-frame information of multiple SDR frames and the contextual information of previously predicted HDR frames to generate high-quality results that are consistent in the temporal domain. The experimental results show that compared with many representative methods, the proposed method can reconstruct higher quality HDR videos.

Pages: 11 to 15

Copyright: Copyright (c) IARIA, 2025

Publication date: March 9, 2025

Published in: conference

ISSN: 2519-8432

ISBN: 978-1-68558-245-6

Location: Lisbon, Portugal

Dates: from March 9, 2025 to March 13, 2025