Research On Image Denoising Combined Grey System Theory And Contourlet Transform

With the rapid development of computer technology, the theory and technologyof image processing has achieved a lot of important results, and the image processingtechnology has been widely applied. The images are often polluted by noise whenpeople get image. The image noises affect the visual effects of the image and makeimage processing difficult to continue. Therefore, image denoising is indispensablestep in the image processing. This paper focuses on the research of grey system theoryand contourlet transform, and has introduced several traditional denoising algorithms.The paper mainly involves the following:(1) The paper described the background and significance of image denoise andintroduced several kinds of common image noise model and several common methodsof image denoising and summarized the status and characteristics of the grey systemtheory and have laid a soild basis for the follow-up study.(2) On the basis of analysis of grey correlation degree and Contourlet transformthreshold denoising algorithm, an improved Contourlet transform denoising algorithmbased on gray relational degree was proposed. On the one hand,considering the grayrelational degree and inter-scale which from the high-frequency sub-band and lowfrequency sub-band by Contourlet transform, the Bayesian threshold is improved; Onthe other hand, in order to achieve the purpose of adaptive denoising, thecharacteristics of Contourlet coefficients is used to improve the compromise thresholdfunction.(3) Concerning the characteristics of single image noise is irregular, an imagedenoising algorithm combining GM (1,1) model and contourlet transform wasproposed. In order to reduce the amount of calculation, the GM (1,1) model isoptimized which is according to the characteristics of image pixel gray value and theGauss noise. A scope was set in order to automatically save strong noise points whichthe GM (1,1) model can’t accurately estimate. In order to achieve the purpose ofdenoising, the forecasted image was disposed by contourlet transform. Consideringthe inter-scale and intra-scale dependencies of contourlet transform and the factor ofnoise intensity, threshold of each subband was set. Finally, contourlet transformcoefficients were processed by threshold function. (4) In order to explore the regularity of single image noise, an image denoisingalgorithm combining GM (1,1) model and wavelet transform was proposed. Thepixel and its neighborhood pixels is modeled by GM (1,1) model, and according thelaw of development of pixel and its neighborhood pixels to update the noise image’spixel. In order to eliminate the negative effect of negative pixels, the pixel value plus1after normalizing the image when model noised image. In order to reduce theamount of calculation, the GM (1,1) model is optimized which is according to thecharacteristics of image pixel gray value and the Gauss noise. A scope was set inorder to automatically save strong noise points which the GM (1,1) model can’taccurately estimate. In order to achieve the purpose of denoising, the forecasted imagewas disposed by wavelet transform after the pixel value being subtracted1. Finally,wavelet transform coefficients were processed by threshold function.