Research On Image Denoising Methods Based On Contourlet Transform

The digital images are gradually becoming a major source of human accessing toinformation. And the process of image acquisition and transmission is ofteninfluenced by external factors, so that the image is polluted by noise. In order toeffectively extract the information from noisy image, it needs the help of imageprocessing technology to reduce noise. This paper has introduced image denoisingalgorithm based on wavelet transform. On this basis, the research of image denoisingalgorithm in Contourlet transform framework has studied as follows:(1) The significance of research background and the subject selection of digitalimage denoising are elaborated, and the development of image denoising fromwavelet transform to multiscale geometric analysis method is introduced. Thetheoretical foundation and structure focused on contourlet transform is discussed,which provides theoretical guidance for further research of image denoising algorithmunder the framework of the contourlet transform.(2) Two key factors of classic threshold denoising algorithm based on wavelettransform are introduced, which are selection of the threshold and threshold function.And the advantages and disadvantages of the different threshold and thresholdfunction are analyzed respectively. In addition, an overview of the basic operations ofthe morphology and the selection rules of structuring element is provided, whichdemonstrates the affect of image denoising in morphological filtering.(3) The frame of Contourlet transform has a very good directional property, but inthe classic threshold denoising algorithm, the threshold processing of all the highfrequency coefficients uses the same of threshold and neglects the correlation ofcontourlet coefficients. Therefore, concerning the defects of multi-scale threshold ondirectional information using Contourlet transform, a new adaptive thresholddenoising algorithm was proposed, which was based on average energy ofneighboring window. According to the distribution of the coefficients energy, theContourlet coefficients were divided up into three areas. The noise could be decreasedobviously by adjusting the threshold of these areas with different variables.(4) While adding the energy of coefficients in neighborhood to thresholdevaluation, although the threshold has considered with neighborhood information ofcontourlet coefficients, the handling capacity of denoising algorithm for high-noise image is still limited. Through analysis and comparison, it proves that usingmorphological filter is an effective method for image denoising. Firstly, the wholealgorithm is parallel consists of two channels. Secondly, the details of the structureelements are added to the structuring elements of morphology. Finally, preprocessedimages with Contourlet transform threshold denoising has fused to the final denoisingimage.