Novel low light image enhancement techniques using foreground background separation

Abstract

Digital image and video processing systems are crucial in numerous newlinefields, including remote sensing, monitoring, intelligent transportation, and newlinesecurity surveillance. The rapid advancement in computer vision technology newlinehas heightened the demand for high-quality visual content. However, images newlineand videos captured under Low-Light (LL) circumstances frequently struggle newlinewith poor contrast and noise due to insufficient exposure and sensor newlinelimitations, thus resulting in degraded visual quality. Enhancing the visibility newlineof LL Images (LLIs) and videos is vital for various high-level tasks, such as newlineObject Detection (OD), recognition, and tracking. Traditional enhancement newlinetechniques often amplify noise and produce biased colors, thereby newlinenecessitating the development of more effective solutions. newlineThis research aims to develop novel Deep Learning (DL)-based newlineframeworks for enhancing LL images and videos. Specifically, the study newlineproposes two optimized enhancement techniques: one for images using newlinePearson Correlation Coefficient induced Golden Jackal Optimizer (PCCGJO) newlineand Mexican ResNet adapted ZFNet (MexResZFNet), and another for videos newlineusing the Bessel Function adapted Retinex-Successive Entropy Quantization newlineTransform (BFR-SEQT) technique. These frameworks are designed to newlineimprove visual quality by addressing the limitations of existing newlinemethodologies. newline