摘要：The work aims to find a novel technique to removenoise from low light or low luminous level images to improvethe visibility of the image and the performance of many imageprocessing systems. A denoising technique using patch priorsin wavelet domain for images with low luminous levels, withthe help of the Gaussian Mixture Model, is presented here.The main idea is to perform denoising in a sparse domain.Initially, the image is decomposed into approximate and detailedcomponents with the help of wavelet transform, and thenthe patch based Gaussian mixture model denoising processis applied on both approximate and detailed components.Expectation maximization algorithm is used for estimating theGaussian mixture model parameters from the image patches.After denoising each component, inverse wavelet transform isapplied to obtain the denoised output image. This denoisingmethod was applied to a set of natural low luminous levelimages, and it resulted in clean images with good Peak Signalto Noise Ratio and Structural Similarity Index, compared toother conventional methods. This work is a novel methodcombining wavelet transform and Gaussian mixture model forthe denoising of low light images.
