Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Frequency Enhancement Network for Single Image Deraining

Fei Yan, Yuhong He, Keyu Chen, En Cheng, Jikang Ma

TL;DR

Single-image deraining is hindered by nonuniform degradation across frequency bands in real rain. The paper introduces AFENet, an end-to-end framework with three modules: FDM for frequency decomposition, FEM for feature enhancement via MDTA and GDFN, and FAM for cross-band aggregation using Pyramid Attention. The approach enables adaptive enhancement across high-, mid-, and low-frequency components, yielding state-of-the-art PSNR/SSIM on Rain13K and RainDS and best no-reference metrics on Real15 and Real300. Experimental results demonstrate AFENet’s effectiveness in removing diverse rain patterns while preserving details, highlighting its potential for practical deployment in rainy-conditions vision tasks.

Abstract

Image deraining aims to improve the visibility of images damaged by rainy conditions, targeting the removal of degradation elements such as rain streaks, raindrops, and rain accumulation. While numerous single image deraining methods have shown promising results in image enhancement within the spatial domain, real-world rain degradation often causes uneven damage across an image's entire frequency spectrum, posing challenges for these methods in enhancing different frequency components. In this paper, we introduce a novel end-to-end Adaptive Frequency Enhancement Network (AFENet) specifically for single image deraining that adaptively enhances images across various frequencies. We employ convolutions of different scales to adaptively decompose image frequency bands, introduce a feature enhancement module to boost the features of different frequency components and present a novel interaction module for interchanging and merging information from various frequency branches. Simultaneously, we propose a feature aggregation module that efficiently and adaptively fuses features from different frequency bands, facilitating enhancements across the entire frequency spectrum. This approach empowers the deraining network to eliminate diverse and complex rainy patterns and to reconstruct image details accurately. Extensive experiments on both real and synthetic scenes demonstrate that our method not only achieves visually appealing enhancement results but also surpasses existing methods in performance.

Adaptive Frequency Enhancement Network for Single Image Deraining

TL;DR

Single-image deraining is hindered by nonuniform degradation across frequency bands in real rain. The paper introduces AFENet, an end-to-end framework with three modules: FDM for frequency decomposition, FEM for feature enhancement via MDTA and GDFN, and FAM for cross-band aggregation using Pyramid Attention. The approach enables adaptive enhancement across high-, mid-, and low-frequency components, yielding state-of-the-art PSNR/SSIM on Rain13K and RainDS and best no-reference metrics on Real15 and Real300. Experimental results demonstrate AFENet’s effectiveness in removing diverse rain patterns while preserving details, highlighting its potential for practical deployment in rainy-conditions vision tasks.

Abstract

Image deraining aims to improve the visibility of images damaged by rainy conditions, targeting the removal of degradation elements such as rain streaks, raindrops, and rain accumulation. While numerous single image deraining methods have shown promising results in image enhancement within the spatial domain, real-world rain degradation often causes uneven damage across an image's entire frequency spectrum, posing challenges for these methods in enhancing different frequency components. In this paper, we introduce a novel end-to-end Adaptive Frequency Enhancement Network (AFENet) specifically for single image deraining that adaptively enhances images across various frequencies. We employ convolutions of different scales to adaptively decompose image frequency bands, introduce a feature enhancement module to boost the features of different frequency components and present a novel interaction module for interchanging and merging information from various frequency branches. Simultaneously, we propose a feature aggregation module that efficiently and adaptively fuses features from different frequency bands, facilitating enhancements across the entire frequency spectrum. This approach empowers the deraining network to eliminate diverse and complex rainy patterns and to reconstruct image details accurately. Extensive experiments on both real and synthetic scenes demonstrate that our method not only achieves visually appealing enhancement results but also surpasses existing methods in performance.
Paper Structure (17 sections, 4 equations, 5 figures, 4 tables)

This paper contains 17 sections, 4 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional Single Image Deraining
  4. Deep Learning-based Single Image Deraining
  5. Proposed Method
  6. Frequency Decomposition Module
  7. Frequency Enhancement Module
  8. Frequency Aggregation Module
  9. Experiments
  10. Implementation Details
  11. Metrics.
  12. Comparisons with State-of-the-art Methods
  13. Datasets
  14. Quantitative Analysis
  15. Qualitative Analysis
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Example of rain removal in rainy conditions.
  • Figure 2: Frequency Analysis of Various Rainy Images (including rain streaks, raindrops, and rain accumulation) and Clean Images: The left side of the analysis highlights the differences in energy distribution among various frequency components. On the right side, the focus is on the variations in Mean Squared Error (MSE) between rainy and clean images across various frequencies. Here, $\Delta$ signifies the absolute difference in rainy degradation across these different frequency components.
  • Figure 3: The architecture of our proposed Adaptive Frequency Enhancement Network (AFENet) includes three key modules: the Frequency Decomposition Module (FDM), which employs a multi-branch architecture to decompose the rainy image content in the frequency domain, thereby enhancing texture details across all frequency components; the Frequency Enhancement Module (FEM), designed to improve the feature representation capabilities; and the Frequency Aggregation Module (FAM), which aggregates the enhanced frequency components from various scales of the rainy image.
  • Figure 4: Frequency Aggregation Module (FAM): a multi-scale feature generation unit to produce features at multiple scales.
  • Figure 5: Visual Comparison with Existing Deraining Methods on the Rain100H and Test100 Datasets. The highlighted red boxes focus on the detailed aspects of the deraining results. Please zoom in for an enhanced view.