Image dehazing via gradient response and bright region adjustment

The presence of atmospheric haze severely compromises image clarity, contrast, and detail, leading to perceptual confusion for humans and adversely affecting the performance and accuracy of computer vision systems. The challenges lie in the difficulty of enhancing the generality and robustness of methods in complex environments. We propose a haze removal algorithm based on gradient response and bright region transmission rate adjustment, specifically addressing the following issues. In contrast to previous methods, we introduce a novel multi-scale fusion technique. Firstly, with the help of Laplace operator, the dark channel gradient matrix at different scales is obtained. By analyzing the image edges and the different degrees of feature changes, we successfully captured the thresholds of strong and weak gradient changes. Using the thresholds as boundaries, we extracted the key scales with significant changes in image features, inverted their gradient strengths and mapped them to fusion weights, which enhanced the details while maintaining the image integrity. Additionally, we present a completely independent adjustment algorithm for transmission rates, effectively avoiding inaccuracies associated with traditional estimation methods. To evaluate the performance of our proposed approach, we employ nine quantitative evaluation metrics on five benchmark datasets. The results demonstrate the outstanding performance and robustness of our method across various complex environments. Its significantly high lower bound further facilitates achieving higher scientific standards.