Low-light Image Enhancement Based on Multi-scale and Mixed Loss Function

Aiming at the problems of low brightness, weak contrast, and blurred edges of low illumination images, a multi-scale and mixed loss function based enhancement algorithm (MMLNet) for low illumination images is proposed. Firstly, the input image is decomposed into the illumination and reflectance, and the multi-scale channel and pixel attention module is designed, which is more conducive to the network learning the mapping relationship between low-light image and clear image. Secondly, the brightness enhancement network is used to correct the illumination of the low-light image to improve its brightness. At the same time, the edge extraction network composed of residual structure is designed to extract clear edge information, and the mixed loss function is added to remove a lot of noise. Finally, the final enhancement result is obtained through image reconstruction. The experimental results show that the proposed algorithm can not only improve the image's visual effect but also PSNR, SSIM and NIQE.