Research On Image Modeling And Image Denoising Of Multiscale Geometry Analysis

Recently, wavelets theory is widely used in image processing and coefficients modeling.Wavelet analysis can be used to effectively describe the signal with point singularities.However, for the signals with singularities distributed along various curves, curve surfaceand hyper-curve-surface in high dimensional space, wavelet analysis loses its advantages,but these characteristics are what we are most interested in. Therfore, wavelet transform isnot the best describing method. In the last several years, a similar theory, multiscalegeometry analysis possesses a better performance in detecting the singularities of the signalwith straight line or curve in high dimensional space, and can be used to express signalswith sparse coefficients.Based on the multiscale geometry analysis theory, we studied a multiscale geometryanalysis tool contourlet transform, which can capture the anisotropic geometricalstructures of real image efficiently. Because of its multiscale and multi-directionalproperty, this paper presented the related coefficients modeling algorithms and applicationsin image denoising. The main work of the dissertation can be summarized as follows:Based on the interscale and intrascale dependencies of the coefficients in ContourletTransform (CT) domain, a contourlet-based Gaussian mixture scale modeling of imagedenoising algorithm is presented. Compared with wavelet-based algorithm, theproposed algorithm is advantage at detail information preservation and the noisesuppression, which is an efficient method of image denoising.Based on real image statistical property andl Bivariate Shrinkage(BivShrink) model,a Coutourlet domain denoise algorithms for image denoising is proposed. Combinedwith the prior estimation of noise signal, the neighboring coefficients of inter-scale andintra-scale are described by BivShrink. In addition, Recursive Cycle Spinning isintroduced into the algorithm for translation invariance. Experiments using plentiful alot of real images indicate that the proposed algorithm outperforms the wavelet-basedBivShrink algorithm in terms of noise reduction as well as image detail preservation,obtaining the higher PSNR values.