Research On Polarization Image Denoising Method Based On Division Of Focal Plane Polarization Camera

Compared with traditional optical imaging and measurement techniques based on information such as light intensity and spectrum,polarization imaging and polarization measurement techniques utilizes the polarization characteristics of light to obtain the polarization information of the target scene to realize target recognition and detection in multi-dimension,which is an efficient optical detection and recognition technology.In particular,the combination of polarization imaging technology and digital processing technology can not only effectively solve the problems that traditional optics cannot solve in the field of target detection and recognition,but broaden the research field and application scenarios of polarization optics.Among various types of polarization imaging systems,the division of focal plane(DoFP)linear polarization imaging system has developed rapidly with its unique advantages and has received extensive attention in various fields.However,the polarization image optimization technology is still in its infancy.Polarized images acquired by DoFP cameras,including degree of linear polarization(Do LP)and angle of polarization(Ao P),are subject to noise.Especially,the Ao P image can be severely degraded under the noise environment.On the other hand,effective denoising technology can improve the quality of polarization images and obtain more accurate polarization information,which has important application value and research significance in practice.In this paper,the polarization image denoising method of DoFP polarization camera is studied.Combined with the structural characteristics and imaging principle of DoFP polarization camera,we proposed a polarization image denoising method based on residual dense neural network,which can make full use of all different levels of feature information and update the network parameters by learning the residual mapping.In this paper,we build several models by different network structure parameters to find the best structure parameters and conduct ablation experiments to verify the effectiveness of the denoising network.The experimental results show that the denoising method can effectively suppress noise of the polarization image and has a good generalization for polarization images of different materials.Compared with existing polarization image denoising methods based on DoFP polarization camera,the proposed method is superior in both quantitative objective evaluation and qualitative visual evaluation,especially denoise on Do LP and Ao P images.The DoFP linear polarization camera can only directly acquire the linearly polarized Stokes vector.In order to further explore the denoising method of the Full Stokes Vector(FSV),we propose two optimization strategies for estimating FSV by matrix decomposition and pseudo-inverse model based on the DoFP polarization camera and the variable phase delay device.These two optimization strategies are aimed at different FSV estimation models,and the measurement variance can be minimized under the condition of two light intensity acquisitions,thereby optimizing accuracy of the measurement.In particular,using the inherent measurement redundancy characteristics of DoFP cameras,we also propose a self-calibration analytical solution for the phase retardation of waveplates based on two intensity acquisitions.This selfcalibration method can replace the process of calibration by phase delay of wave plate in practical applications,while providing technical support for real-time polarization measurement in a dynamic environment.