Back-projection Residual Low-light Image Enhancement Network with Color Correction Matrix

Abstract

Pictures taken in low light conditions suffer from low brightness, feature loss, and color distortion. To address the problem, a Back-projection Residual Network with Color Correction Matrix (BRCCM-Net) for Low-Light Image Enhancement is designed through the following two stages. First, the BRCCM-Net extracts shallow features and inputs into numerous Residual Lightning Back Projection (RLBP) to adjust brightness and learn the residuals of normal light estimations together with the Feature Recalibration module (FRM) to capture and fuse more important feature information, which helps to recover the dark areas of low-light images. Second, a color correction matrix (CCM) model was designed to calibrate luminance and color. The experimental results on the LOL dataset show that the designed network has greater subjective visual effects such as brightness improvement, less noise, color distortion, and sharper and more natural image details. Comparatively, the results for objective indicators (PSNR, SSIM, and MSE) are also superior to those of other existing techniques.