Research On MR Image Rice Noise Removal Algorithm Based On Curvelet Transform

Magnetic resonance imaging is an imaging technique that uses the principle of physical Nuclear Magnetic Resonance(NMR)to make non-invasive observations of human organs,and provides an important means for modern medical diagnosis and treatment techniques.By analyzing nuclear magnetic resonance images,doctors can be assisted to make appropriate diagnoses.However,magnetic resonance images often carry noise during the acquisition process,which reduces the quality of the image.The quality of magnetic resonance images is of great significance for the calculation of relevant parameters for clinical medical treatment and feature information.Therefore,it is very important to reduce the image and improve the image quality,which has become a hot topic in related research.The magnetic resonance image is obtained by sampling and performing the modulo operation of two magnetic resonance signals in the frequency domain.The two signals often carry Gaussian white noise during the acquisition process,resulting in the magnetic resonance image noise after reconstruction being expressed as a Rice distribution.Unlike additive Gaussian noise,the distribution of Rice noise is related to the data of the image,making it difficult to remove.The key to image denoising is to remove background noise and protect important information such as details and edges.Based on this idea,many denoising algorithms have been applied to the denoising processing of magnetic resonance images,such as wavelet transform method,non-local mean denoising algorithm and curvelet transform method.However,these algorithms can easily lose the details of the image while removing the Rice noise.In order to better remove Rice noise,this paper proposes a magnetic resonance image Rice noise removal algorithm based on the new multi-scale transformation theory of curve-let transform.The main research methods are:1.Combine the variance stable transform technique to change the noise in the magnetic resonance image from the Rice distribution to the Gaussian distribution with stable variance.2.The curve-wave transform is performed on the transformed image to obtain the curve-wave domain coefficients,and the denoise is performed by the hard threshold method and the Bayesian soft threshold method,which are respectively recorded as VSTCT-hard method and VSTCT-Bayes method.3.Aiming at the defects of the above two methods in the rough layer of the curvelet transform coefficient without effective denoising,a method of denoising the coarse layer coefficient using soft threshold function is proposed,and it is simple and easy to implement without adding new parameters.The threshold selection method is followed by the improvement of the above two denoising algorithms.The improved algorithms are respectively recorded as VSTCT-hard improved algorithm and VSTCT-Bayes improved algorithm.4.Perform inverse curve transform on the demodulated curvelet coefficients,and then perform variance inverse transform to obtain the final denoised image.In this paper,MATLAB is used as the simulation tool,and the brain nuclear magnetic resonance image provided by Harvard database is used as the original image.The proposed algorithm is simulated and compared with other algorithms.The results show that the proposed algorithm can effectively remove Rice noise,and the evaluation of peak signal-to-noise ratio and average structural similarity is greatly improved,and good denoising effect is achieved.