United Scattering Transmission Model for Haze Removal
Zizhen Wang, Rui Wang, Anran Li, Tingting Ji · Symmetry · 2026
Haze removal methods based on the estimation of scene depth ratio in the Atmospheric Scattering Model (ASM) have achieved satisfactory results. However, the ASM ignores the blur equivalent to a point spread function caused by forward scattering. This paper proposes a simplified United Scattering Transmission Model (USTM), in which both forward scattering and back scattering are taken into consideration physically. It utilizes Taylor expansion to correlate the hazy image and its second-order operator with the dehazed image. Additionally, we establish a layered decomposition mechanism of the scattering medium; by fitting the limitation expression and the image signal at infinity, the parameters related to the inherent optical properties used in the model can be obtained. When the stable transmittance estimation approaches are applied into this USTM, the scene radiance can be restored effectively. We conducted evaluation experiments on datasets including RESIDE-RTTS (Real-world Task-Driven Testing Set), Haze4K, and DenseHaze, using metrics such as PSNR, SSIM, newly visible edges and the ratio of the gradients. The results demonstrate that USTM achieves satisfactory results across multiple evaluation dimensions. Regarding the core objective fidelity metric PSNR, it achieves an optimal score of 11.87 dB, representing an approximate 3.85% improvement over the second-best method. Compared to the traditional ASM, the USTM shows an average improvement of approximately 23.5% in edge restoration capability (newly visible edges) and an average improvement of approximately 18.1% in gradient fidelity (the mean ratio of the gradients). Furthermore, compared with advanced deep learning dehazing methods, our method remains highly competitive in edge and gradient restoration metrics, and its lightweight design provides excellent efficiency and compatibility with downstream tasks. The comprehensive results show that the USTM achieves effective improvements in both physical accuracy and detail restoration performance.