A novel haze removal computing architecture for remote sensing images using multi-scale Retinex technique

Abstract

One of the major degradation arise in the satellite images is poor weather conditions (i.e., haze). Haze is an environmental distortion which significantly affect and minimize the efficiency of remote sensing image interpretation. To tackle this complication, this research work introduced a new approach named as multi-scale retinex histogram equalization with U-Net Dense Optimized Pyramidal (MSRHE with U-Net DOPT) scheme. In this scheme, multi-scale retinex technique is applied to eliminate ambient atmospheric light value of inhomogeneous information and retain only object surface reflection but they suffer from dark tone. So, of contrast limited adaptive histogram equalization is employed to brighten the image tone with limited contrast level. Then the enhanced image is restored using U-Net Dense Optimized Pyramidal technique which deeply extract the feature information with dragonfly optimized ReLU activation unit to enhance the system efficiency. This resolved criteria can greatly improve the image efficiency to generate haze-free image. At the same time, the techniques such as dark channel prior (DCP) method, urban remote sensing haze removal (URSHR) method were employed to comparatively analysed with MSRHE with U-Net DOPT. Three parameters such as peak signal to noise ratio (PSNR), feature similarity (FSIM) and structural similarity index measure (SSIM) were employed to quantitatively estimate the test (query) result. The experimental result concludes that the proposed technique works well when compared with other classic approaches.