Research On Adaptive Threshold Image Denoising Algorithm Based On NSCT Domain

With the advent of the information age,images have became the main way for people to obtain information.When images are collected,stored,and transmitted,they may be affected by noise for various reasons,causing people to misunderstand the image contents and adversely affect subsequent image processing.Therefore,image denoising has became an important image preprocessing technique.Image denoising is the removal of doped noise information from the image.Its purpose is to enhance image quality and highlight the image itself.Therefore,image denoising has an indispensable role and position in the field of image engineering.Non-subsampled Contourlet transform(NSCT transform)is one of the typical denoising algorithms.The denoising algorithm based on NSCT transform has the characteristics of multi-resolution,multi-directionality,anisotropy,translation invariance and so on,which can be able to describe the image information better.Although the NSCT transform has the advantages that denoising algorithms should have,there are some deficiencies in threshold processing and parameter estimation,and there is space for improvement and optimization.In this thesis,the above problems are studied and corresponding improvement algorithms are proposed to improve the effect of denoising.The main research contents are as follows:Firstly,aiming at the problem of ringing and pseudo-Gibbs caused by the discontinuity of hard threshold function in the NSCT domain,and the problem of image blur distortion caused by constant value compression in the soft threshold function,an image denoising algorithm is proposed based on improved threshold function in NSCT domain.The improved threshold function introduces the idea of exponential smoothing function algorithm to make it continuous,gradual,deviating and high-order conductibility in the NSCT domain,overcomes the existing problems of soft and hard threshold functions.In this algorithm,the BayesShrink adaptive threshold estimation is used in the threshold estimation part,which can determine the threshold size accurately and solve the phenomenon that traditional fixed threshold estimation excessive remove the transform coefficient.Secondly,for the wavelet transform in the wavelet-domain Hidden Markov Tree model can not well capture the insufficiency of image geometric features,a denoising statistical algorithm based on the NSCT Hidden Markov Tree model(HMT model)is presented.Because the NSCT transform can describe the characteristics of the transform coefficients in different spaces,scales,and directions,making the HMT model more accurate in parameter estimation.This algorithm also improves and optimizes the HMT model.The state of the coefficient child nodes should not only take into account the influence of its parent nodes,but also take into account the influence of its parent nodes' neighbor nodes.The improved HMT model can more effectively reflect the complex relationship between NSCT coefficients.Finally,the denoising algorithm is tested by comparing experiments in this thesis.The experimental results show that the improved image denoising algorithm proposed in this thesis has better denoising effect compared with the traditional denoising algorithms in condition of objective evaluation criteria such as peak signal-to-noise ratio,root-mean-square error and image enhancement factor.