The Research On Image Information Processing Methods Under Dynamic Imaging Environment

It is an important guarantee to analyze and study the inherent information of the image from high-contrast visualization images,and it is a technical support for quantitatively evaluating motion characteristics of targets to estimate motion parameters from finite-length image sequence.However,in the environment of weak lighting,cloud and backlight,the color and near infrared images are dark and the contrast is extremely low,which greatly hinders the human visual system from identifying the components of low-quality images,even it is impossible to detect and track these targets from these low-quality images.And by the impact of imaging techniques,muscle tissue and dose,classification and extremely strong toxicity of contrast agent,finite-length X-ray angiograms contain additive Gaussian white noise and the cardiovascular contrast is very low,which not only hinders clinicians from diagnosing and analyzing cardiovascular diseases,but also increases the difficulty of cardiovascular extraction,cardiovascular motion analysis and three-dimensional cardiovascular tree reconstruction.To this end,it is necessary to design image information processing methods to improve the image quality and to separate each component from mixed-signals of finite-length X-ray image sequence that can help us construct an efficient target monitoring and intelligent analysis system.In this dissertation,some key issues of image information processing methods are studied and discussed to deal with the above problems with the support of the National Projects,such as studying and analyzing the efficiency and reliability of these methods which can be used to support the basic research on target monitoring and intelligent analysis system.The research results and contributions of this dissertation are as follows:First,two different denoising methods are proposed,including:(1)a iterative weigted nuclear norm(IWNN)denoising method,a new surrogate function is proposed to solve the problem of the big error of the nuclear norm(NN)approximation to the rank norm,then the prior NN problem is converted into a IWNN problem and an alternative direction iteration method is employed to get the approximate solution of the IWNN problem.(2)A spatially adaptive image denoising method is proposed.As the dual-domain filter is greatly affected by the gradient factor,a new spatially adaptive gradient factor is proposed based on the characteristics of eigenvalues of the Hessian matrix.Lots of results of experiments on synthetic images demonstrate the efficiency of the two proposed denoising methods.Extensive experiments on X-ray angiographic images further illustrate that the two methods performs well on noise reduction and capillaries preservation,which can provide sufficient information for clinicians to diagnose potential cardiovascular diseases.Next,a novel Hessian-based non-local weighted filter is proposed.In order to enhance cardiovascular and cerebrovascular while removing noise and preserving vascular structures including vascular edges,capillaries and peripheral vessels in X-ray angiographic images,a non-local weighted factor is introduced into the Hessian matrix to obtain noiseless eigenvalues.By the use of these eigenvalues and the similarity of local pixels,a novel enhancement method,which can preserve vascular structures,is proposed.Experimental results show that this method can effectively suppress non-vascular areas and noise whiling maintaining vascular tree structures(especially capillaries and vascular terminals).Again,two different methods are proposed to obtain adaptive Gamma parameters.As the prior Gamma parameter cannot adjust the brightness of images,this dissertation firstly correct the probability density function with global and local cumulative histograms,then the functions are normalized to get corrected cumulative histograms which are used to construct adaptive Gamma parameters.Lots of experimental results show that two adaptive Gamma correction methods for enhancing near-infrared images can effectively improve the contrast of near-infrared images,maintain their high brightness and preserve structure details.Then,a variational enhancement method based on Retinex with framelet regularization is proposed.The framelet transform has the ability of capturing multi-scale structures,the main idea of this method is to preserve multi-scale structure details of reflectance by the framelet regularization to reflectance.The two methods including alternative direction iteration and splitting Bregman iteration are employed to simultaneously estimate reflectance and illumination which are beneficial to improve color image quality.Experimental results illustrate that the proposed method performs well on improving the brightness of low-contrast images,enhancing their contrast and preserving their structure details,which greatly enhances their visibility.Finally,an optimal time-frequency domains iteration separation method is proposed.As the traditional Fourier transform can not accurately separate the finite-length mixed-signal,this dissertation separate the multi-motion parameters of the baranch feature points of cardiovascular by using the alternative minimum criterion of global mean square error in the time domain and the local mean error in the frequency domain.The results of a large number of simulated and real mixed-signal separation experiments show that the proposed method can effectively separate the multi-motion parameters of the finite-length mixed-signal of branch feature points,which can provide technical supports for clinicians to quantitatively analyze and diagnose cardiovascular diseases.