Single Image Dehazing Using Multi-Scale DCP-BCP Fusion

Dehazing involves finding the solution to an ill-posed problem. Improvement to the traditional prior based dehazing method has been introduced here. Considering the pivotal role of an accurate transmission map estimate in dehazed image restoration, a fusion based dehazing algorithm is proposed, where refined transmission maps from different scales are fused to obtain an overall transmission map estimate, with fewer block artifacts (owing to selection of a single patch size for a single scale model). At each scale, transmission is estimated using both weighted dark channel(DCP) and bright channel priors(BCP), effectively avoiding color distortions. Additionally, Weighted Dark channel and a Cross-Bilateral filter for refinement, helps prevent halo artifacts. The model has been compared with other sophisticated models using both ground truth and non-ground truth metrics. Experiments show that our algorithm can effectively diminish haze, while retaining the color of pure scene radiance, clearly evident in sky region. MATLAB has been used for simulation and analysis purposes.