Low-light Image Enhancement Algorithm Based on Curve Estimation and Denoising
Under the interference of low light,backlight and non-uniform light,it is difficult to obtain high-quality image en-hancement.To relieve the issue caused by the problem mentioned above,this paper proposes a low-light image enhancement algo-rithm based on curve fitting and denoising.The network is mainly composed of four sub networks:curve estimation,decomposition,denoising and optimization.Furthermore,the network is supervised by the perceptual loss and detail loss,which is the main loss function.Afterwards,a result is obtained by the decoder module,which has a better performance in contrast,detail,color,noise and so on.Moreover,a residual learning module is employed to make the deep neural network more easier to optimize in the training stage,and will alleviate the problems that caused by the gradient disappearing or explosion.The experimental results on LOL datas-et show that the algorithm achieves better results than other compared solutions in the metrics of peak signal-to-noise ratio and structural similarity.Compared with our baseline algorithm Zero-DCE,the proposed method has better performance on both metrics of PSNR and SSIM,which have huge gains of 14.7%and 32.8%,respectively.
国家科技期刊平台
NSTL
万方数据
