Fast Image Desmogging for Road Safety Using Optimized Dark Channel Prior
Akshay Juneja, Vijay Kumar, Sunil Kumar Singla · 2023
Images captured under dense smog suffer from various problems such as distortion of spatial information and poor visibility. The majority of existing restoration models are for other variants of weather degradation techniques such as fog, haze, rain and dust. It is challenging to create a restoration model for smoggy images as they affect the machine vision systems such as aerial imaging and sensing imaging. In this chapter, a novel deep learning-based desmogging architecture is proposed. The transmission map and atmospheric veil are estimated using optimized dark channel prior (DCP). The result generated by DCP is satisfactory for images distorted due to foggy or hazy weather conditions, but because of the presence of black fumes due to smog gradients in the atmosphere it is difficult to separate sky and non-sky regions. To overcome the disadvantage, the transmission map of the image obtained after computation using DCP is optimized by enhancing contrast and entropy of the image. Various performance metrics are calculated and a comparative analysis of the proposed model with existing desmogging or dehazing algorithms is presented.