Low-Light Image Enhancement Combined with Attention Map and U-Net Network

Abstract

The quality of images obtained in low-light environments will degrade and exhibit poor results, so researchers have focused on improving low-light images and enhancing them. This paper recommends a different method for low-illumination image enhancement. Our method combines traditional methods and CNN to achieve image brightness enhancement, color recovery and denoising. The attention map that we propose can help avoid overexposure in bright regions and enhance dark regions. Through a large number of subjective and objective experiments, the method proposed in this paper can not only enhance low-illuminance images, but also require less inference time.