Medical Ct Image Denoising Algorithm Based On Discrete Non-separable Shearlet

Medicine has become one of the most important subjects in the history of human development,and CT imaging provides great help for medical diagnosis,whereas it is inevitable that noise will be generated when collecting data of human tissue,which will affect the diagnosis of the disease.Therefore,the study of CT denoising is of great significance.The prior study found that the essence of CT image denoising is to utilize the noisy image to estimate the noise-free image of human organ tissue,and obtain the optimal estimation of the real image,which has important practical significance for diagnostic accuracy.Therefore,we propose a medical CT denoising algorithm based on discrete and non-separable Shearlet transform by analyzing the principle of CT imaging and the distribution characteristics of noise in the transform domain in this paper.Compared with the traditional wavelet algorithm,the algorithm has more directionality in the frequency domain,which can preserve the detailed features of the image.Medical CT images are decomposed into high-frequency and low-frequency parts by multi-scale and multi-directional transformation through DNST,and obtain the optimal distribution model by statistically analyzing the distribution of shear wave coefficients.An improved threshold shrinkage algorithm is proposed,and the algorithm can combine the difference of each coefficient in high-frequent sub-band to generate different optimal thresholds.Combined with the characteristics of the low-frequency part,a hybrid low-frequency filter is proposed.The processed coefficients in high frequency sub-band and low frequency sub-band coefficients are reconstructed by inverse discrete non-separable shearlet transform to obtain denoised medical CT images.In order to verify the superiority of the proposed algorithm,a large number of comparative experiments,including synthetic simulation image experiments and medical CT image experiments,are carried out.Based on the subjective and objective evaluation,the proposed algorithm performs well in denoising effect and can be applied well in medical real-time imaging systems.The main work and results of this paper are as follows:(1)According to the characteristics and statistical peculiarity of the sub-bands of the noise in the transform domain,the threshold function is improved to calculate the optimal threshold of each coefficient in high-frequency sub-band;(2)Shearlet transform solves the limitation of the direction and isotropic of wavelet transform.On this basis,the discrete non-separable shearlet transform is proposed.This transform has a better frame boundary than the traditional shearlet transform,and the sub-band directionality in the frequency domain is improved by the sector filter.(3)Based on the soft threshold shrinkage algorithm and the hard threshold algorithm,the OLI contraction algorithm is proposed based on the characteristics of the image;(4)A hybrid low-frequency filter is proposed.The filter combines guided filtering,bilateral filtering and rotational invariant bilateral non-local mean filtering.The algorithm uses the iterative feature to combine the advantages of the three algorithms to ensure the balance between denoising effect and detail features;(5)In the experimental part,we combined the experimental results data with various evaluation indicators for subjective evaluation and objective analysis by analyzing the contrast experiments and clinical medical image experiments.