The Research On Image Denoising Using Wavelet Transform, Dependency And Edge Preservation

Image is an important information source. However, in the course of its acquisition,transmission and storage, noise is often introduced, which degrades the quality of image.So image denoising is an important task of digital image processing. A comprehensiveresearch on image denoising using wavelet transform, dependency and edge preservationis made in this dissertation, which mainly includes the following contents:First of all, detailed introduction of wavelet analysis development in recent years andits application areas, especially image denoising area are given in this paper, and thenthree kinds of present conventional image denoising methods are listed. Moreover, theprinciples and characteristics of these methods are analyzed, too.On the basis of these analyses, three methods are proposed, that is, the method ofwavelet image denoising based on neighbour dependency and adaptive thresholding, themethod of wavelet image denoising based on multiscale edge detection and adaptivethresholding and the method of nondecimated wavelet image adaptive denoising withedge preservation. The first method focuses on image denoising without existing Gibbsphenomena. In order to construct the soft threshold function, a circular neighbourhoodaround the wavelet coefficient to be thresholded is considered, whist it combines with adata-driven and adaptive threshold which is derived from Bayesian estimation. The resttwo methods stress removing noises with edge preservation. And the idea about"subdivide wavelet coefficients" is proposed. At each resolution level, we make adifference between the wavelet coefficients related to edge, the coefficients associated tohomogeneous regions and the coefficients noise-related. Then these coefficients whichhave dissimilar traits are processed respectively with different strategies.At last, the methods for image denoising proposed in this paper are testedrespectively. Experimental results show that these methods often obtain better denoisingperformance than the existing typical methods.