Visibility Enhancement in Fire and Rescue Operations: ARMS Extension with Gaussian Estimation

In fire and emergency rescue operations, visibility is often severely degraded by smoke, airborne debris, or atmospheric pollutants including smog and yellow dust. Several image restoration techniques, including Dark Channel Prior (DCP), Color Attribution Prior (CAP), Peplography, and Adaptive Removal via Mask for Scatter (ARMS), have been proposed to recover clear images under such conditions. However, these methods exhibit significant limitations in heavy scattering environments. This paper proposes a novel visibility restoration method for disaster situations, building upon the state-of-the-art ARMS method. To maximize the suppression of scattering effects, the Scattering Media Model is refined through Gaussian estimation. Additionally, an overlapping matrix is introduced to effectively handle non-uniformly distributed scattering conditions. The proposed method is evaluated using a real rescue operation image dataset provided by the Fire and Disaster Management Agency of Japan. Qualitative visual assessments and quantitative performance metrics demonstrate that the proposed approach significantly outperforms conventional methods under severe scattering conditions.