Frequency-consistent Optimization for Image Enhancement Networks

Bing Li (University of Science and Technology of China),* Naishan Zheng (University of Science and Technology of China), Qi Zhu (University of Science and Technology of China), Jie Huang (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
The 34th British Machine Vision Conference


Image enhancement aims at enhancing the overall contrast (low frequency) while reconstructing details (high frequency). Existing studies typically achieve these two objectives with a heuristically constructed complex architecture (i.e., two-stage or two-branch). In contrast, we attempt to perform the image enhancement task within a single-stage and single-branch network. However, directly employing a single plain network to optimize the two objectives simultaneously will lead to an optimization conflict between contrast enhancement and texture restoration, resulting in suboptimal performances. To alleviate this problem, we construct a frequency-independent feature space for maintaining optimization consistency. Specifically, we propose a Frequency Decorrelation and Integration (FDI) module with two core insights: 1) formulating a frequency-independent space via decorrelation normalization to bridge the frequency discrimination; 2) integrating the initial frequency-dependent features with a channel shuffle operation for information complement and reducing the sensitivity to frequency during optimization. Therefore, these two designs encourage networks to learn along the optimization direction of frequency consistency. In addition, the proposed FDI is a plug-and-play module that can be incorporated into the existing methods with negligible parameters. Extensive experiments on various image enhancement benchmarks demonstrate consistent performance gains by utilizing our proposed module.



author    = {Bing Li and Naishan Zheng and Qi Zhu and Jie Huang and Feng Zhao},
title     = {Frequency-consistent Optimization for Image Enhancement Networks},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year      = {2023},
url       = {}

Copyright © 2023 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection