SASM-DOS: construction of an atmospheric scattering model for smoke environments and an algorithm for smoke image clarification

Given the serious image degradation, complex noise, and non-uniform illumination in smoke environments, this paper analyzes the imaging characteristics of smoke environments, combines the diversity of smoke media components and scattering characteristics, constructs an atmospheric scattering model for non-uniform smoke environments, and proposes an image clarification algorithm for smoke environments accordingly. The proposed smoke environment image clarification algorithm employs a multi-stage clarification framework. First, a multi-scale hierarchical denoising strategy based on a pyramid structure is proposed for the noise distribution characteristics of smoke images, suppressing noise layer by layer while retaining details. Second, to address non-uniform illumination, an atmospheric light estimation method based on superpixel segmentation (SLIC) is designed, using cross-bilateral filtering to eliminate the chunking effect and improve the local consistency of the atmospheric light matrix. Subsequently, a joint transmittance estimation method based on scene radiance and saturation-intensity components is proposed to enhance edge features and suppress transmittance estimation bias. Finally, to alleviate color deviation and low contrast in smoke images, the preliminary restoration results are converted to the LMS color space. Color fidelity and contrast are optimized using channel linear stretching, while the grayscale dynamic range is enlarged. Experimental results show that the proposed smoke image clarification algorithm improves the detail texture and color fidelity of smoke images in a subjective visual assessment. Compared with other smoke environment image processing methods, it performs better in all quantitative evaluation metrics. The proposed method can provide robust technical support for scenarios such as fire rescue, industrial monitoring, and autonomous driving.