AoSRNet: All-in-One Scene Recovery Networks via Multi-knowledge Integration

TL;DR

AoSRNet tackles the challenge of restoring visibility in diverse low-visibility imaging conditions by fusing physics-informed preprocessing (gamma correction and optimized linear stretching) with learning-based refinement. The architecture combines GC-guided detail enhancement (DEM), OLS-guided color restoration (CRM), a multi-receptive field extractor (MEM), and an Encoder-Decoder fusion module (EDFM) built on standard residual blocks to robustly recover scenes across haze, sand-dust, and low-light environments. Key contributions include the multi-knowledge integration strategy, the MEM for texture preservation under nonlinear/linear transforms, and a loss suite that combines L1, color similarity, and contrastive regularization. Empirical results on RESIDE-OTS, RESIDE-SOTS, and SMD show AoSRNet achieves superior or competitive performance with strong generalization, and the approach offers practical utility for vision-driven urban, automotive, and robotic systems; source code is publicly available.

Abstract

Scattering and attenuation of light in no-homogeneous imaging media or inconsistent light intensity will cause insufficient contrast and color distortion in the collected images, which limits the developments such as vision-driven smart urban, autonomous vehicles, and intelligent robots. In this paper, we propose an all-in-one scene recovery network via multi-knowledge integration (termed AoSRNet) to improve the visibility of imaging devices in typical low-visibility imaging scenes (e.g., haze, sand dust, and low light). It combines gamma correction (GC) and optimized linear stretching (OLS) to create the detail enhancement module (DEM) and color restoration module (CRM). Additionally, we suggest a multi-receptive field extraction module (MEM) to attenuate the loss of image texture details caused by GC nonlinear and OLS linear transformations. Finally, we refine the coarse features generated by DEM, CRM, and MEM through Encoder-Decoder to generate the final restored image. Comprehensive experimental results demonstrate the effectiveness and stability of AoSRNet compared to other state-of-the-art methods. The source code is available at \url{https://github.com/LouisYuxuLu/AoSRNet}.

Figures (9)

• Figure 1: Example of the scene recovery in three different imaging conditions. The upper triangles in (a)-(c) are degraded patterns, and the corresponding restored patterns by our method are shown in the lower triangles.
• Figure 2: The flowchart of the all-in-one scene recovery network (AoSRNet). It mainly contains gamma correction (GC)-guided detail enhancement module (DEM), optimized linear stretching (OLS)-guided color restoration module (CRM), multi-receptive field (MRF) extraction module (MEM), and Encoder-Decoder-based fusion module (EDFM). Standard residual block (SRB) is the basic unit of learning.
• Figure 3: Suggested basic composition of (a) Convolutional Layer (ConvL) and (b) standard residual block (SRB).
• Figure 4: Processed results of gamma correction operations of (a) hazy and (b) low-light image with different $\gamma$ values (i.e., $\gamma=\frac{1}{4},\frac{1}{2},1,2$, and $4$).
• Figure 5: The atmospheric light value and its corresponding three-channel histogram distribution of the two types of degraded scenes.