Enhancement of Hazy and Low Light Images Using Hybrid Restoration Methods and Deep Learning Techniques

Abstract

Images captured in low light conditions or in hazy weather often suffer newlinefrom reduced visibility, degraded contrast, and poor visual quality, which can newlinenegatively impact various applications such as surveillance, autonomous driving, and newlinemedical imaging. Hybrid and deep learning algorithms have recently emerged as newlineeffective methods for improving the visual quality of these images. This research work newlineproposes novel methods that combine hybrid techniques, such as dual channel prior newlineand Retinex-based methods, with deep learning techniques, including Convolutional newlineNeural Network (CNN), for efficient enhancement of low light and hazy images. newlineThe first proposed method combines Dual Channel Prior (DCP) and newlineRetinex-based techniques to eliminate haze in both daytime and nighttime images. The newlineDCP method utilizes the brightness and saturation channels of an image to estimate newlinethe atmospheric light and transmission map, while the Retinex-based method enhances newlinethe contrast and brightness of the image. The two methods are integrated in a hybrid newlineframework, where the estimated transmission map is used to adaptively apply the newlineRetinex-based enhancement to different regions of the image. This hybrid method newlinetakes advantage of the strengths of both techniques, effectively removing haze from newlineimages captured in both daytime and nighttime conditions newline