Research On Full Polarization Parameters Estimation And Polarization Imaging Contrast Optimization Based On Statistical Properties Of Noise

Polarization information is weakly correlated with traditional intensity information and spectral information.Polarization imaging thus shows great advantages in tasks that cannot be achieved by applying intensity or spectral imaging.Effective use of polarization information can enhance imaging performance,improve target detection and recognition rate.Therefore,it can be applied in domains such as target detection,navigation,3D reconstruction,remote sensing,and biomedical diagnosis.However,the existence of noise degrades its acquisition precision and destroys the final effect of its application.Therefore,the investigation of the optimal strategy for polarization information acquisition and application in the presence of noise with different statistical properties is the key issue in polarization research.Full polarization parameters estimation and polarization imaging contrast optimization are typical representatives of polarization information acquisition and application respectively.When noise exists,the noise modeling method and polarization imaging technology involved in their optimization research process can be directly applied to other polarization information acquisition and application.The research on them provide theoretical support and technical guidance for high-precision acquisition and effective utilization of polarization information in the presence of noise.Therefore,it is of great significance to study them.In this thesis,we focus on the interference of noise with different statistical properties on the acquisition and application of polarization information.Taking noise as the main line,research on full polarization parameters estimation and polarization imaging contrast optimization are carried out.The main work is as follows:(1)The polarization information of a light beam can be completely represented by full polarization parameters,including Stokes vector,degree of polarization(DOP),angle of polarization(AOP),and ellipticity of polarization(EOP).Most polarization applications are based on the acquisition of these parameters.DOP,AOP and EOP are nonlinear functions of Stokes vector.Stokes vector can be measured by devices called"Stokes polarimeters".During their acquisition process,there are two main types of perturbations:additive Gaussian noise(AGN)and Poisson shot noise(PSN),which lead to the acquired parameter values deviate from their true values.However,there is a lack of complete theory to construct the relationship among statistical properties of noise,the polarization state of the measured light,the measurement matrix of the Stokes polarimeter and the acquisition accuracy.Aiming at this problem,the models of full polarization parameters estimation are established according to the statistical properties of noise.Then,based on Delta method,an analytical but approximate method is proposed to derive the variances of DOP,AOP and EOP.Moreover,we derive analytical expressions of estimation variance of full polarization parameters for arbitrary measurement matrices in the presence of AGN and PSN.These expressions are validated through Monte Carlo simulations and optical experiments.These analytical expressions provide a clear insight into how signal to noise ratio,the polarization state of the measured light,and the measurement matrix govern the acquisition accuracy of full polarization parameters.These expressions can be used as optimization criterions to optimize the measurement matrix to achieve optimal estimate of full polarization parameters.For a particular Stokes polarimeter,these expressions can be used to describe acquisition precision of full polarization parameters acquired by Stokes polarimeter on a quantitative basis,which provide evidence of the reliability level for the application of full polarization parameters.(2)The above estimation variances of DOP,AOP and EOP are obtained by an approximation method based on Delta method.However,these expressions have a limited domain of validity.By comparing the results of the approximation method with those of Monte Carlo method,we determine the validity domain of the approximations to mean value and variance of the estimator of DOP,AOP and EOP,when the measurement matrix is spherical-based and noise is AGN.We demonstrate that the domain of validity is quite large and the approximated variances fairly accurately reflect the actual estimation precision.At their point of non-differentiability and in the presence of strong noise,DOP and EOP approximations show however some limits.Aiming at these problems,we thus propose other approximations(based on statistical properties of estimators)with extended domain of validity that are more mathematically complex and have no analytical solution.One can selectively use these two approximations according to the application requirements.This study provides evidence of the validity domain for the application of the theory in(1).(3)Polarization imaging contrast optimization based on full polarization parameters estimation is one of the most important applications of polarization information,whose purpose is target detection.The existence of noise destroys the performance of contrast optimization.Choosing three typical scenes,using an active polarization system with Quasi-monostatic configuration under varying scene orientation angle,through the analysis of theory and experiment,we investigate polarization imaging contrast optimization in the presence of noise.The results reveal that if only Gaussian and Poisson noise are considered for contrast optimization,the obtained contrast is often insufficient.This is because in real-world images,another type of noise is dominant:spatial fluctuations of polarization properties of the scene.Moreover,the law of illumination and analysis polarization state that can achieve the optimal contrast are summarized for real-world polarization imaging detection.The results are important for optimizing polarization imager for target detection.