Objective In hazy weather conditions, images captured by conventional cameras often suffer from blurring and distortion due to the effects of particle scattering. Image dehazing has emerged as a focal research area to achieve clear imaging under these adverse conditions. This technique is primarily employed to enhance visual clarity, improve image visibility, and facilitate optical analysis in specialized environments. It finds extensive applications in remote sensing, target recognition, defense and military sectors, as well as autonomous driving technologies. Polarization imaging technology, as an emerging optical-electronic imaging technique, captures both intensity and polarization information, providing distinct advantages in extracting target shapes and texture details, which has garnered widespread attention in the field of dehazing. Existing polarization dehazing algorithms often require the extraction of sky regions, a process that is intricate and prone to introducing artifacts. Moreover, the calculated polarization degree and polarization angle images are significantly affected by particle scattering effects, leading to substantial information loss. To address the aforementioned challenges, this paper proposes a fusion dehazing algorithm based on polarization dark channel.

Methods A fusion dehazing algorithm utilizing polarization information is built in this paper. Based on the principle of dark channel prior, the polarization dark channel is extracted to estimate the atmospheric light and transmission rate for obtaining the scene image. Subsequently, a global adaptive tone mapping method is employed to restore the image brightness. Additionally, hazy images and ground truth images were collected through artificial hazing experiments. An analysis of the effects of haze scattering revealed the issue of blurred image details. Therefore, a local histogram equalization method is introduced to enhance image contrast. Finally, the results of polarization dehazing were fused with those obtained from local histogram equalization using wavelet transform. The high-frequency components were fused using local saliency fusion, while the low-frequency components were fused using weighted average fusion. Various image evaluation metrics are then computed to assess the effectiveness of the algorithm.

Results and Discussions The images used for algorithm validation consist of 40 sets of real hazy scenes and 20 sets of simulated hazy scenes. The contrast algorithms selected include the Retinex enhancement algorithm, WANG's polarization dehazing algorithm based on dark channel prior, and deep learning-based methods C2Pnet and D-former. Subjectively, the algorithm proposed in this paper achieves high-quality dehazing for both near and distant scenes, with significant detail recovery and superior visual image quality. Objectively, the proposed algorithm shows a significant improvement in contrast. The enhancements in average gradient and local standard deviation are also notable, indicating that the algorithm effectively reduces blur and noise and enhances edges and details. There is also a noticeable increase in information entropy, further reflecting the recovery of image complexity and richness. In summary, the algorithm presented in this paper demonstrates significant effectiveness and superiority, meeting the requirements for clear imaging in hazy conditions.

Conclusions A wavelet fusion dehazing algorithm based on the polarization dark channel prior is proposed. The algorithm has a simple structure, does not require sky region extraction, and is applicable to a broad range of scenarios. Additionally, it provides stable dehazing effects and produces high-quality images. A global adaptive tone mapping method is introduced to solve the image degradation problem. Through the design of a haze imaging experiment, the impact of haze on the grayscale distribution of ground truth images is analyzed. Consequently, the local histogram equalization method is introduced to enhance image contrast. Finally, by employing wavelet image fusion and integrating the advantages of polarization dark channel and local histogram equalization, fusion dehazing images with rich detail recovery and a high dynamic range are obtained. The experimental results indicate that the polarization dehazing algorithm proposed in this paper can effectively restore real hazy weather scenes.