Research On Image Denoising And Fusion Algorithm In The New Types Of Wavelet Transforms Domain

Images are often corrupted by diversified noises during image acquisition and transmission,the objective of denoising is to obtain an estimated image as similar as possible with the clearimage; Image fusion is eager to get a composite image which integrates the complementary andredundant information of multiple images data of the same scene, so that the composite imagecontains better and more accurate description of the scene. Image denoising and fusion are twopreprocessing of image processing, which have an important influence on the subsequent imageprocessing methods such as edge detection, texture analysis, feature extraction and patternrecognition, etc. Therefore, the research of image denoising and fusion is obviously of muchsignificance.Wavelet analysis is good at time-frequency localization, and is widely applied in the fieldof image processing. However, the traditional wavelet transform is shift sensitive and lacksdirection selectivity. To overcome the disadvantages of the traditional wavelet transform, somenew type of sparse representation tools that can better and more sparsely represent image andunstable signal are proposed in recent years. Image denoising and fusion algorithms based onthe new type of wavelet transforms domain not only effectively overcame the spectrumdistortion in spatial domain but also have obtained more ideal results than the traditionalwavelet transform domain. Therefore, it has become a research focus.In this thesis, some new types of wavelet transforms are preliminarily studied, then theimage denoising and fusion applications based on these new types of wavelet transforms arein-depth studied. The main innovations are summarized as follows:1. We studies deeply on the relate theory of quaternion wavelet transform (QWT) and theapplication in image denoising. According to the characteristics of QWT coefficients’magnitude distribution, the generalized Gauss distribution is used to model, an adaptivethreshold is obtained under the Bayesian theory framework. Experimental results show thatour method is not only better than many of classic denoising methods in the peaksignal-to-noise ratio (PSNR), but also obtain better visual effect.2. A novel image denoising algorithm based on nonsubsampled Contourlet transform domainis proposed. First, according to the correlation of nonsubsampled Contourlet transformcoefficients, a classification standard is presented, and the divided coefficients are modeled. Adaptive threshold is derived under the Bayesian theory framework and the best range ofthe parameter is found out. In order to overcome the shortcoming of the soft and hardthreshold function, then a new adjustably adaptive threshold function is presented.Experimental results show that the proposed algorithm outperforms other currentoutstanding algorithms in peak signal-to-noise ratio, structural similarity and visual quality.3. A new image denoising algorithm based on trivariate prior model in nonsubsampleddual-tree complex contourlet transform (NSDTCT) domain and non-local means filter(NLMF) in spatial domain is presented. Firstly, according to the correlation of NSDTCTcoefficients, the distribution of the high frequency coefficients is modeled with the trivariatenon-Gaussian distribution model. A nonlinear trivariate shrinkage function is derived, andinverse NSDTCT is performed to get the initial denoised image. Finally, NLMF is used tosmooth the initial denoised image. Experimental results show that our algorithm can makebetter ideal effect than any other methods.4. A novel adaptive image fusion algorithm based on nonsubsampled shearlet transform(NSST) is proposed. For the low frequency sub-band coefficients, the singular valuedecomposition method in the gradient domain is used to estimate the local structureinformation of image, and an adaptive ‘weighted averaging’ fusion rule based on thesigmoid function and the extracted features is presented. A novel Sum-modified-Laplacian(NSML) is employed in multi-scale product domain to select bandpass sub-bandcoefficients. The proposed fusion method is verified on several sets of multi-source images,and the results show that the proposed approach can not only restrain the noise influence,but also significantly outperform the conventional image fusion methods in terms of bothobjective evaluation criteria and visual quality.