The Research On Light Field Reconstruction Algorithms From Focal Stack Based On Inverse Problem

The spatial and directional information of rays can be recorded by using light field imaging technology,which can realize depth of field extension,digital refocusing,and3 D reconstruction.It is widely applied in the research fields of computer vision and computational imaging.The light field can be obtained directly through devices such as light field cameras or multi-camera arrays.However,due to the large volume,high cost,and difficult calibration of traditional multi-camera imaging,and the trade-off of spatial resolution and angular resolution of light field cameras,how to realize high-quality light field data from a limited data has become a hot research topic in the field of computational imaging in recent years.We focus on the algorithm of how to reconstruct the high-precision light field based on focal stack data.The main contributions and innovations are listed as follows:(1)Under the condition of 3D assumption and infinite slope ρ ∈(0,+∞),we derived the analytical reconstruction formula of the light field,and we established the filtered back-projection(FBP)algorithm from the focal stack.Furthermore,we proved the convergence of our algorithm under continuous points.Since in real data sampling,the light field can only be reconstructed by a small number of focal stack,the problem is a seriously incomplete problem.In order to improve the reconstruction results,a deconvolution algorithm based on FBP method was introduced.The experimental results show that the deconvolution algorithm can effectively improve the quality of the reconstructed light field,and the proposed algorithm performs better with the window which is smooth at the boundary and with larger depth range.(2)Since fast guided filtering has the advantage of denoising while preserving image edge features and relaxation strategy of the iterative algorithm can optimize the reconstruction results,we proposed a filtered-based Landweber iterative method.Simulation data and real data experimental results show that the proposed method can effectively remove the edge noise,and the proposed method is more practical and effective compared with some of the relevant reconstruction methods.(3)The algorithm can be more well-posed by using total variation(TV)regularization term in the iterative algorithm.The inversion of the discrete light field imaging forward model was carried out.We proposed a total variation(TV)regularization sparse model with the alternating direction method of multipliers(ADMM)based on guidedfiltering.Specifically,the updated image in the iteration step contains the guidance image,and the initializer for the least squares method using a least square QR factorisation(LSQR)algorithm was involved in one of the subproblems.In simulated and real data experiments,the model outperforms other existing seven methods in both visual assessments and objective metrics(such as peak signal to noise ratio,PSNR and structure similarity index measure,SSIM).We also show a further application for refocusing by using the reconstructed light field.