Kinder: Dark light image enhancement method based on improved U-net

Aiming at the problems of poor visibility, color distortion, detail loss and low contrast degradation of images acquired in dark light environment, a dark light image enhancement method based on improved U-net is proposed, which is called Kinder. On the basis of Retinex theory, the network is divided into three parts: layer decomposition network, reflectance recovery network and illumination regulation network. Based on U-net, the reflectance recovery network introduces Res-Block and spatial attention mechanism. The addition of Res-Block not only deepens the network structure, but also improves the network’s ability to express features. The spatial attention mechanism could get more attention for low-light area and better solve the degradation phenomenon. The experimental results show that Kinder performs well in objective evaluation index compared with the current advanced dark light image enhancement algorithms, such as Retinex-net, MBLLEN, GLADNET and KinD. Compared with KinD, the SSIM value has increased by 1% and the PSNR value has increased by 20%. This method not only improves the overexposure problem of the enhanced image, but also maintains the color fidelity and details of the image, which has a better performance effect in visual presentation.