| Title |
Wavelet-Based All-in-One Dual-Branch Adverse Weather Image Restoration with Bidirectional Subband Interaction |
| Authors |
박수연(Park Su Yeon) ; 엄일규(Eom Il Kyu) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.8.61 |
| Keywords |
Adverse weather image restoration; All-in-one deweathering; Wavelet transform; Dual-branch network; Bidirectional subband interaction; Recurrent transformer |
| Abstract |
Images captured under adverse weather conditions suffer from degradations caused by rain, snow, and haze, which not only reduce visual quality but also degrade the performance of vision systems. Existing all-in-one adverse weather restoration methods can handle multiple weather degradations within a single network, but they often fail to fully reflect the fact that different weather phenomena affect low-frequency and high-frequency image components in different ways. To address this issue, we propose a wavelet-based all-in-one dual-branch adverse weather image restoration network. The proposed network first decomposes the input image into low-frequency and high-frequency subbands through a wavelet transform, and then extracts and reconstructs restoration features using a dual-branch encoder-decoder architecture specialized for each subband. The low-frequency branch is designed to recover complementary background information, while the high-frequency branch detects adverse weather degradation patterns. In addition, bidirectional subband interaction is introduced so that low-frequency background information guides high-frequency degradation detection, while high-frequency degradation features are fed back to refine low-frequency background restoration. Furthermore, convolutional neural networks and transformers are asymmetrically arranged in the encoders and decoders according to the characteristics of each frequency band, allowing the strengths of both modules to be effectively exploited. Experimental results demonstrate that the proposed method achieves superior restoration performance compared with existing state-of-the-art methods under various adverse weather conditions. |