Deep Residual Convolutional Network for Natural Image Denoising and Brightness Enhancement

Because of the low-light shooting environment, the camera sensor will loss huge details and fuzzy edge. A deep low-light residual convolutional network (LRCNN) is proposed in this paper, which utilizes the sparse coding feature to get the true signal and adaptively adjusts the image exposure in the low-light state. The residual connections in LRCNN help us preserve more potential detail information in the original picture and accelerate the training speed of the network. Many existing image enhancement algorithms only are able to address one aspect of image problems. We designed a neural network system which could deal with many image processing problems at the same time. The experimental results show that our neural network system well optimizes the images that affected by darkness and noise. It also avoids an artificial appearance in generating the image patches.