Research On The Algorithms For High ISO Image Denoising

CCD imaging devices now widely used in military monitoring, community surveillance, civilian photography and other fields. CCD Sensitivity (ISO) as an optical imaging equipment parameter, describes the degree of sensitivity to capture light. Higher ISO setting will enhance freezing of scene motion in dim light condition, because exposure time can be shortened sharply. But since affected by the random Gaussian noise, image's visual appearance will be severely degraded.This paper categorizes the CCD noise firstly, and discusses the reasons for the occurrence of high-ISO noise, then selects CIE-Lab colour space as filtering space. Secondly we present three filters which balance the performance between noise smooth and edge reservation in algorithm design. In order to improve the robustness of Hybrid Mean and Adaptive Center Weighted Median Filter, the noise estimation method based on wavelet decomposition is adopted, and is added the pre-processing of illumination classification to solve the noise distribution inequality problem which is caused by illumination distribution inequality. Within modified wiener filter in wavelet domain, we adopt an infinite-order continuously derivable threshold function and propose a simple and effective iterative stopping criterion which improves the visual fidelity as well as shortening the iteration time. For the purpose to remove block effect caused by classic P-M second-order model, a coupled PDE method for image denoising is proposed based on the P-M model and the fourth-order PDE, which is embedded a Gaussian smoothing kernel to accelerate the diffusion speed in isolated noises.Experiments based on standard test images and high ISO images have shown that the wiener filter in wavelet domain suppresses noise apparently as well as preserves fine details, and processing time is much shorter compared with other two algorithms. We can obtain satisfactory filtered images evaluated both by the objective criteria such as PSNR, MSE, Smooth evaluation index and subjective visual fidelity.