Digital Image Enhancement Using MirNet and Zero-Deep Curve Estimation (Zero-DCE)

Increasing public concern about security due to the lack of CCTV monitoring and lighting at night. The problem is that the quality of CCTV is low if the room does not have normal lighting or even tends to be dark. Images produced from cameras that have poor performance produce dark images so that the resulting information cannot be used for the next process. In this research, the image improvement results of the MIRNet and Zero-DCE methods were compared, from this comparison the best method was obtained. To improve the quality of low/dark images, deep learning-based approaches such as MIRNet and Zero-Deep Curve Estimation (Zero-DCE) have received great attention. This study aims to compare the performance of both approaches in the context of image enhancement in low-light conditions. This research uses Low-Light (LoL) and real CCTV datasets. The data is divided into training and testing data. Then the data was analyzed using MIRNet and Zero-DCE. The test results using MIRNet have a PSNR of 27.86 dB and an MSE of 10.595. Meanwhile, Zero-DCE has average PSNR and MSE values of 28.055 dB and 10.204. The PSNR and MSE results show that Zero-DCE has the best average results between the two existing methods.