Low-Light Image Enhancement Network based on Gamma Correction and Multi-scale Attention Mechanism

A low-light image enhancement method based on Gamma correction and a multi-scale attention mechanism is proposed for the problem of detail loss due to low image brightness and contrast under low-light conditions. First, we extract the image luminance and color feature maps through an attention network, construct Gamma values using each pixel of the feature maps, and perform Gamma correction on the input low-light images to enhance the brightness and contrast of the images. A multi-scale attention network is designed to extract the noise distribution features of the low-light image, and a U-Net network is used to achieve image noise suppression. Experiments show that compared with other existing algorithms, the method in this paper has some improvement in the objective indicators of low-light image enhancement, PSNR, SSIM and NIQE.