Abstract

 Images captured at night with low-light conditions frequently have a loss of visible details, inadequate contrast, low brightness, and noise. Therefore, it is difficult to perceive, extract, and analyze important visual information from these images, unless they were properly processed. Different algorithms exist to process nighttime images, yet most of these algorithms are highly complex, generate processing artifacts, over-smooth the images, or do not improve the illumination adequately. Thus, the single scale retinex (SSR) algorithm is adopted in this study to provide better processing for nighttime images. The proposed algorithm starts by converting the color image from the RGB model to the HSV model and enhancing the V channel only while preserving the H and S channels. Then, it determined the image’s illuminated version somewhat like the SSR, computes the logarithms of the illuminated and original images, then subtracts these two images by utilizing an altered procedure. Next, a modified gamma-adjusted Rayleigh distribution function is applied, and its outcome is processed once more by an automatic linear contrast stretching approach to produce the processed V channel that will be utilized with the preserved H and S channels to generate the output RGB image. The developed algorithm is assessed using a real dataset of nighttime images, evaluated using three dedicated image evaluation methods, and compared to ten dissimilar contemporary algorithms. The obtained results demonstrated that the proposed algorithm can significantly improve the perceptual quality of nighttime images and suppress artifact generation rapidly and efficiently, in addition to showing the ability to surpass the performance of different existing algorithms subjectively and objectively.