Research On Partial Differential Equation Noise Reduction Model For Medical Ultrasound Images

Medical image assistant and processing technology is an important auxiliary tool of modern medical diagnosis,which plays an important role in medical diagnosis.Medical ultrasound diagnosis technology has become one of the most widely used auxiliary tools in clinical diagnosis,because of its real-time,safety and low price.However,due to the inherent characteristics of ultrasonic imaging equipment,the ultrasonic image is often affected by noise,which not only reduces the image quality,conceals the important details of the image,but also affects the accuracy of subsequent advanced image processing.The ultrasound image with clear edge details is an important basis for doctors to diagnose the disease.Image noise reduction model based on partial differential equation has the advantage of excellent noise reduction ability and solid mathematical theory.Thus,it has been widely studied and applied.The image denoising model based on partial differential equation can be divided into second order partial differential equation denoising model,fourth order differential equation denoising model and fractional differential equation denoising model according to the differential order.In order to improve the noise reduction ability and edge preserving ability of the traditional partial differential equation denoising model,this paper studies the traditional second-order,fourthorder,fractional order partial differential equation denoising model and the joint application of partial differential equation denoising model and other denoising algorithms.This paper is divided into the following four parts:1.By studying the image denoising model based on second-order partial differential equation,we propose two new denoising models based on second-order partial differential equation.1)In order to improve the edge preserving ability of second-order partial differential equation denoising model,an image denoising model based on adaptive weighted average algorithm is proposed.Firstly,the principle of image average is applied to preprocess the image to reduce the image noise;secondly,by analyzing the advantages of the classical PM model and SRAD model,an adaptive weighted average algorithm based on PM model and SRAD model is designed.According to the principle of statistics,the edge information of the denoised image is extracted.Adaptive weighted average algorithm is used to enhance the edge of preprocessed image.The noise in the final denoised image is smoothed effectively and the edge is clear.The experimental analysis shows that the model is simple in principle,fast in calculation,and takes into account the ability of noise smoothness and edge preservation.2)In order to overcome the problem that the second-order partial differential equation noise reduction model is prone to "gradient effect",a new model of second-order partial differential equation noise reduction based on block gradient is proposed by optimizing the diffusion function threshold to control the diffusion velocity in different regions.Firstly,the block gradient is used to calculate the edge eigenvalue of image,which overcomes the influence of noise on the edge.Secondly,the block gradient operator is used as the edge detection operator instead of the gradient model value in the original model,and a new noise reduction model is constructed.The experimental results show that the model based on block gradient can not only smooth the noise,but also improve the ability of image slope edge preservation.2.In order to solve the problem that the traditional fourth-order model is easily affected by isolated noise and the noise reduction level is low,a fourth-order partial differential equation noise reduction model coupled with gradient fidelity term and regularization term is proposed.Firstly,the improved mean filter is used to preprocess the image,and the second-order degree is calculated according to the preprocessed image,which overcomes the problem that the traditional fourth-order model is sensitive to isolated noise;secondly,the gradient fidelity term introduced in the model can control the model to perform weak smoothing operation on the image edge,so as to retain the image edge information;the regularization term can reduce the image noise in the model The detail information is lost in the iteration to keep the authenticity of the image.Compared with the original model,the new model not only improves the ability of noise smoothing and edge preserving,but also avoids the influence of isolated noise.3.In order to solve the problem that the traditional fourth-order model is easily affected by isolated noise and the noise reduction level is low,a fourth-order partial differential equation noise reduction model coupled with gradient fidelity term and regularization term is proposed.Firstly,the fourth-order partial differential equation model proposed in Chapter 3 is extended to fractional order,and the obtained fractional order partial differential equation denoising model has the advantages of both fourth-order and fractional order denoising models;a mean square curvature calculation method is designed to improve the ability of the model to retain texture features and small-scale information;at the same time,the numerical calculation method and implementation steps of the new denoising model are given.Compared with the traditional fractional order noise reduction model,the new fractional order model has better noise reduction ability,boundary preserving ability and detail processing ability.4.Analyzing the advantages and disadvantages of PDE denoising model and other denoising algorithms,two denoising algorithms are proposed: PDE denoising method combined with wavelet transform;PDE denoising method combined with non-local mean.1)The noise reduction method based on partial differential equation combined with wavelet transform firstly decomposes the image into low-frequency and high-frequency components through wavelet transform.Since most of the noise is concentrated in high-frequency components,the new method uses low-frequency components to calculate the edge indicator function,which can reduce the influence of noise on edge features.Secondly,the improved PM model is used to analyze the different frequency wavelet components of the image adaptively.Finally,different frequency components are integrated to restore the denoised image.The new method combines the advantages of wavelet transform and partial differential equation denoising,suppresses the noise from the perspective of frequency,and has strong universality.2)Combined with the partial differential equation denoising method of nonlocal mean,the concepts and calculation methods of nonlocal gradient and nonlocal divergence are extended by using the concept of non-local mean.In this method,the non-local feature gradient is used to replace the traditional gradient value,which improves the image internal similarity;the iterative non local mean denoising method,which combines the iterative denoising characteristics of the partial differential equation denoising method and the non-local mean denoising method,improves the robustness of the new method to noise.