H.5. Image Processing and Computer Vision
Ali Shabani Badi; Kambiz Rahbar; Ziaeddin Beheshtifard; Maryam Khademi
Abstract
This paper introduces a novel approach to enhance the quality of images captured under low-light conditions. The method optimizes the parameters of the established Li method by employing the evolutionary Particle Swarm Optimization (PSO) algorithm. A key contribution of this research is the formulation ...
Read More
This paper introduces a novel approach to enhance the quality of images captured under low-light conditions. The method optimizes the parameters of the established Li method by employing the evolutionary Particle Swarm Optimization (PSO) algorithm. A key contribution of this research is the formulation of a comprehensive loss function for the PSO algorithm, derived from the integration of entropy loss, edge pixel loss, and average desired image brightness loss. The objective of this optimization process is to determine the optimal parameter set for the base method, thereby improving the preservation of image structure, increasing brightness while maintaining edge details, and ensuring the overall brightness of the resulting image remains within a desirable range. An iterative optimization strategy is employed to address the resulting optimization problem. The performance of the proposed method is evaluated through quantitative and qualitative analyses on the SICE dataset and benchmarked against several state-of-the-art low-light image enhancement techniques. Quantitative evaluation, utilizing metrics such as PSNR, SSIM, PIQE, NIQE, BRISQUE, and NIMA, demonstrates that the proposed parameter tuning of the base method, guided by the PSO algorithm and our comprehensive loss function, achieves competitive or superior performance in preserving image structure and details, generating images with natural visual quality, and suppressing noise in comparison to numerous existing methods. This research highlights the efficacy of the evolutionary PSO algorithm in identifying optimal configurations for a physical model-based method aimed at enhancing the quality of low-light imagery.
