Several Filtering Algorithms For Removing Mixed Noises In Digital Image

Many kinds of noises will be introduced into image during acquisition and trans-mission, which will lead to image quality decline. In application, there are two kindsof noise models mainly represent most noises which added to images: Gaussian noiseand impulse noise, of cause, their mixed noises are inevitably. Removing the mixednoise is important and challenging subject in image processing.Recently, there is much work carried out on removing mixed noises. In 2005,Guan Xin-ping and others proposed removing mixed noise filter. In 2005, R Garnettand others proposed Trilateral Filter. Additionally, in 2007, Liu Quansheng and Li-bing and others proposed the Mixed noises Non-local Filter, MNF for short, and soon.This paper mainly study two kinds of algorithms of removing mixed noises, whichis Gaussian noise mixed with uniform impulse noise:One is based on the Trilateral Filter, using linear weight factor replace non-linearweight factor, and a new Linear Mixed Filter is proposed, LMF for short. The algorithmis faster than Trilateral Filter and increased about 15%; moreover, both visual e?ectand PSNR have increased di?erently. Further, based on the LMF, studied more intwo facts: On the one hand, using ROLD replaces ROAD in detecting impulse noise,and proposed ROLD Linear Mixed Filter, RLMF for short. The algorithm increasedthe ability of removing uniform impulse noise and mixed noise. On the other hand,gradient can enhance the image detail and edge, in the weight function, using theneighbor pixel's gradient similarity replaces the pixel's value similarity. A GradientLinear Mixed Filter is proposed, which based on the pixel structure, GLMF for short.The algorithm e?ectively protects image details and edges in some degree.The other is based on the MNF, using two similar neighbor window pixels'meanvalue rate and the variance rate as the threshold, an accelerating algorithm aboutMNF proposed, which based on pixel's similarity (Fast-MNF), FMNF for short. Thealgorithm is faster than MNF and increased about 10%, and the denoising e?ect isnot worse than MNF. Besides, using the image texture and details with directional,an accelerating algorithm about MNF is proposed, DMNF for short. The algorithm isfaster than MNF and increased about 25%, and the denoising e?ect is not worse thanMNF.This paper mainly includes six chapters. The first chapter introduces the digitalimage processing, the significance of image denoises, and the image denoises studyactuality. The second chapter introduces the digital image, image noise, and image denoises theory basic. The third chapter introduces sorts of filter method which wereproposed nearly a decade. The fourth chapter proposed sorts of new linear mixed noisefilter methods. The fifth chapter introduces two accelerating algorithm about MNF:FMNF and DMNF. The sixth chapter summarized this paper's work, and proposedthe coming study direction.