Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rethinking the Elementary Function Fusion for Single-Image Dehazing

Yesian Rohn

TL;DR

This work targets the limitations of purely physical or purely data-driven dehazing methods by introducing CL2S, a DM2F-based network that substitutes the logarithmic component with a sinusoidal model to better capture complex haze distributions. The architecture fuses five elementary-function components through learned attention, preserving image details and color while handling inhomogeneous haze. Extensive ablations validate the necessity and effectiveness of each component, and CL2S demonstrates strong performance across RESIDE, O-HAZE, HazeRD, and C-Haze datasets. The approach offers a practical path toward robust dehazing by expanding the function space used to model haze and integrating physical principles with learnable modules.

Abstract

This paper addresses the limitations of physical models in the current field of image dehazing by proposing an innovative dehazing network (CL2S). Building on the DM2F model, it identifies issues in its ablation experiments and replaces the original logarithmic function model with a trigonometric (sine) model. This substitution aims to better fit the complex and variable distribution of haze. The approach also integrates the atmospheric scattering model and other elementary functions to enhance dehazing performance. Experimental results demonstrate that CL2S achieves outstanding performance on multiple dehazing datasets, particularly in maintaining image details and color authenticity. Additionally, systematic ablation experiments supplementing DM2F validate the concerns raised about DM2F and confirm the necessity and effectiveness of the functional components in the proposed CL2S model. Our code is available at \url{https://github.com/YesianRohn/CL2S}, where the corresponding pre-trained models can also be accessed.

Rethinking the Elementary Function Fusion for Single-Image Dehazing

TL;DR

This work targets the limitations of purely physical or purely data-driven dehazing methods by introducing CL2S, a DM2F-based network that substitutes the logarithmic component with a sinusoidal model to better capture complex haze distributions. The architecture fuses five elementary-function components through learned attention, preserving image details and color while handling inhomogeneous haze. Extensive ablations validate the necessity and effectiveness of each component, and CL2S demonstrates strong performance across RESIDE, O-HAZE, HazeRD, and C-Haze datasets. The approach offers a practical path toward robust dehazing by expanding the function space used to model haze and integrating physical principles with learnable modules.

Abstract

This paper addresses the limitations of physical models in the current field of image dehazing by proposing an innovative dehazing network (CL2S). Building on the DM2F model, it identifies issues in its ablation experiments and replaces the original logarithmic function model with a trigonometric (sine) model. This substitution aims to better fit the complex and variable distribution of haze. The approach also integrates the atmospheric scattering model and other elementary functions to enhance dehazing performance. Experimental results demonstrate that CL2S achieves outstanding performance on multiple dehazing datasets, particularly in maintaining image details and color authenticity. Additionally, systematic ablation experiments supplementing DM2F validate the concerns raised about DM2F and confirm the necessity and effectiveness of the functional components in the proposed CL2S model. Our code is available at \url{https://github.com/YesianRohn/CL2S}, where the corresponding pre-trained models can also be accessed.
Paper Structure (15 sections, 8 equations, 3 figures, 2 tables)

This paper contains 15 sections, 8 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background and Significance
  3. Challenges in Dehazing
  4. Physical Models and Prior Knowledge
  5. Deep Learning and Dataset Bias
  6. Inhomogeneous Haze Distribution and Feature Channel Differences
  7. Related Work and Research
  8. Method
  9. Review of Baseline Method
  10. Elementary Function Fitting
  11. Experimental Results and Analysis
  12. Datasets
  13. Implementation Details
  14. Ablation Study
  15. Conclusion and Future Work

Figures (3)

  • Figure 1: CL2S architecture, with specific modifications replacing $J_{4}$ in the bottom right corner.
  • Figure 2: Examples of CL2S and DM2F performance on mainstream datasets
  • Figure 3: Performance of CL2S and DM2F on C-Haze