Efficient Non-Local Similarity-based Image Dehazing: A Pixel-Level Approach for Enhanced Performance and Robustness

This research study proposes an efficient approach for image dehazing using non-local similarity-based methods. Dehazing has roots in many applications, including image enhancement of underwater imagery, satellite aerial images, remote sensing imagery, and many others. The proposed method considers the challenges of single-image dehazing. The method involves the identification and recognition of haze lines. After identifying haze lines, a regularization process is introduced to consider the variance of the estimated haze lines. This regularization ensures that only the pixels adhering to the model's assumptions contribute to the reconstruction process. The proposed algorithm works at the pixel level rather than at the patch or regions of patch-level techniques. This results in improved speed, robustness, and reduced sensitivity to parameters like patch size and content. Performance analysis was done using several focus measures. Experimental results demonstrate the effectiveness of the proposed method. The simulation results prove that the proposed method outperforms the state-of-the-art methods.