Research On Several Key Technology Of Computational Imaging For Light Field Display

As a research hotspot of 3D display technology,light field display technology realizes the reproduction of light field in the scene through the light control element,and provides dense and continuous viewpoint information,meanwhile,it puts forward higher requirements for the acquisition of 3D visual information.The sparse light field content provides fewer viewing angles and lower resolution of a single viewpoint,while the dense light field content has a high sampling density which leads that the light field display presents a more continuous and realistic scene.Through the light field computational imaging technology,the captured sparse light field can be reconstructed into a dense light field,which meets the requirements of high-quality light field 3D display.Light field computational imaging technology can be divided into light field acquisition stage and subsequent calculation processing stage.Existing light field acquisition systems such as camera arrays and light field cameras can realize multi-viewpoint acquisition of scenes,but due to the constraints of hardware and cost,the light field content obtained directly through acquisition cannot meet the both requirements of high spatial resolution and high angular resolution.In addition,high-quality light field content requires high manpower and hardware costs,and urgently requires a matching copyright protection method.The above problems have been researched in this paper.The specific research content and innovation points of the paper are as follows:(1)A light field computational imaging method based on external masks and compressed sensingIn order to meet the needs of light field display,in terms of light field acquisition,a sparse light field acquisition method is proposed,the acquisition system combines an external mask and a common camera,and then the compress sensing theory is used to reconstruct the target light field.The designed external mask is fixed in front of the lens,and the light field with 25 viewpoints is reconstructed from images which are obtained by taking multiple shots with different apertures.Firstly,we analyze the light field in the common camera which is combined with an external mask.In order to solve the problem of uneven brightness of sub-aperture images caused by vignetting effect,a brightness correction method is proposed.In order to obtain the measurement matrix corresponding to the external mask,a measurement matrix estimation method based on threshold judgment is proposed.Using the K-SVD dictionary learning method,a light field dictionary containing 3444 atoms is trained,which realized the sparse representation of the light field.The simulation results show that the image structure similarity of each viewpoint in the reconstructed light field obtained by this method is at a high level,the SSIM values of all viewpoints are above 0.8,and some viewpoints are above 0.9.The reconstructed light field of the real scene has achieved good results in applications of light field display and digital refocusing.(2)A depth estimation method based on axially distributed light fieldThe depth image based rendering technology can realize the dense reconstruction of the light field,and the accuracy of this method depends on the accuracy of the depth image.Compared with the traditional parallel distributed light field,the axial distributed light field is easy to obtain,the hardware cost is low,and it can provide dense viewpoint information with higher spatial resolution.Based on this,the research on depth estimation of light field is carried out in this paper,and a depth estimation method for axially distributed light field is proposed.A color similarity score function for initial depth estimation is constructed from the analysis of the axially distributed light field.In order to optimize the initial depth results,a confidence level function is constructed according to the color similarity score distribution curve.An energy optimization model is established using credibility function.The simulation results show that under the sampling density of 100 viewpoints,the PSNR of the depth estimation results of some scenes can reach 30dB.Compared with the depth estimation method based on depth tomographic images matching,the depth estimation result of this method is significantly improved.An axial acquisition system for real scenes is built,combined with the depth estimation method and DIBR algorithm proposed in this paper,multi-view light field content is generated,and the effect of light field display proves the effectiveness of the method.(3)A digital watermark encryption method for light field contentThe copyright protection problem of light field images needs to be solved urgently.The traditional watermark encryption method for 2D images is studied in this paper,and a digital watermark encryption method for light field images combined with light field theory is proposed.To match the dimensions of the light field content,a method for extending the 2D watermark to light field is proposed.For the watermark coverage area,a set of DWT-SVD based watermark embedding and extraction rules are formulated.The depth tampered light field content will affect the watermark extraction results,to solve this problem,a watermark matching detection algorithm based on depth traversal is proposed.The experimental results show that the visual change of the carrier light field is small after embedding the watermark,and the PSNR corresponding to each viewpoint image and the refocusing image is above 50dB,which proves that the method has strong invisibility.Under some conventional watermark attack methods,the extracted watermark has a high similarity with the embedded watermark,the corresponding normalized similarity coefficient is maintained around 0.9,and the QR code watermark information can be effectively identified,which proves that the method has strong robustness.In the case where the light field content is tampered with the depth,the watermark matching detection method successfully matches the watermark,and the detected parallax offset is the same as the tampered parallax offset,which confirms the effectiveness of the method.