Research On Low-light Image Denoising For The Smartphone Picture

In recent years,the shooting function of smart phones has become increasingly powerful,and the resolution of mobile phone photos is constantly increasing.Due to the limitation of the size of image sensors in mobile phones,the area of photosensitive cells is continuously reduced,which makes the signal-to-noise ratio of pictures decrease,especially under low-light condition.Studying low-light image denoising technology,restoring image details and improving image quality has important research significance and can be widely used in consumer electronics,security monitoring and autonomous driving.The paper expounds the research background and significance of low-light image denoising technology,and summarizes the research status of related technologies.Using the time domain information of multi-frame images,a multi-frame fusion denoising algorithm based on continuous shooting images is proposed.Firstly,the paper analyzes that low-light images have serious noise interference and low contrast.Aiming at the shortcomings of optical flow sensitivity to noise,an improved SIFTflow motion estimation method is proposed,which combines dense SIFT maps and the structure map to improve the registration accuracy of the algorithm under noise.Experiments show that the proposed registration algorithm can not only obtain dense motion vector information,but also be more robust to noise.Secondly,in view of the failure of registration,the paper proposes a detail-conserved image fusion denoising algorithm;Using the transform domain fusion method to maintain the merged image details.Using a consistent pixel map to reduce the noise residual based on the transform domain fusion method,and an adaptive weighted fusion method is designed to further reduce the image noise after fusion.The experimental results show that the proposed fusion algorithm can better preserve the image details while removing noise.Based on the Cross-channel Correlation characteristics of image noise,a single-frame image denoising algorithm based on Cross-channel noise model is proposed.The paper analyzes the sources of Cross-channel correlation characteristics of image noise and introduces the cross-channel noise model.Based on the non-local self-similarity and low-rank prior hypothesis of images,the cross-channel Weighted Nuclear Norm is derived based on the Cross-Channel noise model and used for image denoising.The inverse matrix of the covariance matrix in the used in model solving process is not semi-positive definite because of the inaccuracy of estimation.A semidefinite matrix is used to approximate the original matrix seek to the solution of the Nearest Correlation Matrix Problem.The experimental results show that the proposed image denoising algorithm outperforms the state-of-the-art denoising methods...