Abstract:

Atmospheric Phenomenons like haze, fog, mist, dart are produced due to tiny suspended particles floating in the air which scatter, reflect, refract, absorb light in all directions spherically.   As a result , natural images captured under the influence of such turbid weather are prone  to produce degraded visibility leading to fatal accidents in computer vi-sion applications.Single image visibility  techniques are  the most challenging of all models in visibility improvement and solely obey the restoration based optical image formation model. Existing single image dehazing estimates atmospheric light (AL) and transmission by some prior information manu-ally. In this paper, we experimented  the robustness of the Lévy distribution of Cuckoo Search Algorithm(CSA)  in tuning a new  multi-objective image performance  function PPS for adaptable dehazing which estimates and achieves balance between correlation ,noise, and geometrical information of image . Two parameters, AL   and  Depth Map (DM),   are optimised  with levy steps in the search space of CSA  to produce the best dehaze image . GT O-Haze, DerainNet, Frida synthetic dataset have been used for evaluations and  the presented method is compared with  the state-of-the art techniques qualitatively and quantitatively for dehazing obtaining  satisfactory results .

Keywords: