Research On Denoising Algorithm Of Color Polarization Image Based On Focal Plane

Light has four main physical properties: intensity,wavelength,coherence,and polarization.Compared with traditional imaging systems based on the intensity and wavelength of light,polarization imaging systems can provide additional information by using the polarization characteristics of light.The color polarization imaging system utilizes the polarization,wavelength,and intensity characteristics of light to provide a finer and richer polarization imaging effect of red,green,and blue at the same time.It has a wide application prospect in the fields of target recognition and detection,biomedical imaging,three-dimensional object reconstruction and so on.Color focal plane polarization imaging system is a key research direction in the field of color polarization imaging because of its small volume and fast imaging speed.However,the Color focal plane polarization imaging system is composed of image sensors with Color polarization filter array(CPFA).The noise generated in the process of image acquisition may lead to incorrect estimation of color polarization information.After the color polarization interpolation algorithm is used to obtain the color polarization image of the CPFA image,the Stokes vector,and the Degree of linear polarization(Do LP)are calculated.In this process,noise will be further amplified,leading to the loss of structural features and edge details in the polarization information of the object.Therefore,in order to obtain more accurate and richer color polarization information,it is necessary to carry out denoising processing in color polarization imaging system.In this paper,we mainly study the denoising algorithm of CPFA image before interpolation and color polarization image after interpolation.The specific research content includes the following aspects:Firstly,we study CPFA image denoising algorithm(CPFA-PCA)based on Principle component analysis(PCA).In this algorithm,a variable block composed of four polarization channels and three-color channels is established,and spatially similar samples are found to construct a similar block matrix.The eigenvalue matrix and eigenvector matrix of the similar block matrix are used for denoising in the principal component analysis domain.The residual noise of Stokes vector in each color channel is estimated and denoised according to the standard error propagation technique.Finally,Do LP color images reconstructed from simulated and real CPFA images are used to evaluate the performance of the denoising method.The experimental results show that the PSNR value and SSIM value are increased to 1.7 times and 3 times respectively compared with the noisy image.The algorithm can effectively suppress the noise and retain the polarization information,but the image is too smooth.Then,we study a CPFA image denoising algorithm(CPFA-BM3D)based on Blockmatching and 3-D filtering(BM3D).The algorithm can make full use of the correlation between the four polarization directions and the three-color channels,and use the spatial structure characteristics of CPFA image and the correlation between the polarization channel and the color channel to find similar blocks to form 3D blocks,and then carry out3 D transformation and collaborative filtering.The experimental results show that the PSNR value and SSIM value are increased to 2 times and 4 times,respectively,compared with the noisy image.This method has remarkable noise suppression performance while preserving the image details and polarization information.Finally,we study a color Polarization image denoising algorithm based on polarization total variation regularization(PSSTV).Firstly,the 3D structure of the color polarization image is decomposed by low-rank tensor,and the total variational regularization of spacespectrum-polarization is proposed by using the correlation between polarization channel and spectral channel.Then a guide map is generated according to the characteristics of the color polarization image.Combined with the guide map,a weighted spatial difference operator based on polarization information is designed to represent the piecewise smooth structure of the space,spectral domain,and polarization domain.The experimental results show that the PSNR value and SSIM value are 2.1 times and 4.2 times,respectively,compared with the noisy image.This method can suppress the noise of color polarization image well,and retain the details of the image spectrum and polarization information.