Light Field Depth Information Extraction And Rendering Quality

Recently,the development of light field imaging technology has broken the restrictions in traditional cameras,and also brought new ideas to computer vision and virtual reality technology.Light field imaging technology is applied to areas such as virtual reality,autonomous driving,and face recognition,where depth estimation techniques are critical.Depth estimation is a technique that can analyze objects in the scene and understand objects.We can use depth information to interpolate and render new perspective images.For 3 degree of freedom plus video,the depth estimation result of any viewing angle is equally important,and the result determines the quality of the rendered image and thus affects the user's experience.Most of the current algorithms study the depth estimation of the central sub-aperture image and do not consider the accuracy of the non-central sub-aperture images depth estimation.Therefore,it is necessary to study the depth estimation algorithm applicable to arbitrary sub-aperture images.The accuracy of the depth map is not only affected by the depth estimation algorithm,but also the quality of the light field data.In the actual situation,the amount of light field data is very large.If the original light field is directly transmitted,the bandwidth requirement is very high.Therefore,the light field data is typically compressed prior to transmission.From the perspective of objective data differences and subjective users,the influence of light field quality on depth estimation and rendering results is analyzed.The analysis results have important guiding significance for algorithm optimization.Based on this background,this paper proposes a depth estimation algorithm for arbitrary angle sub-aperture images,and designs and implements a compressed light field rendering system to perform subjective and objective quality analysis on the system.In order to improve the accuracy of all sub-aperture images depth estimation in the light field,a rotating parallelogram algorithm with arbitrary sub-aperture image is proposed.The rotated parallelogram algorithm of the arbitrary sub-aperture image detects occlusion based on whether the imaging of the angular sample image is consistent.The non-occlusion point is directly filled with the result of the corresponding central sub-aperture image,and the occlusion pixel is obtained by the rotation parallelogram operator of the adaptive rotation center.On this basis,the credibility matrix is designed to measure the histogram distance obtained from the polar plane images in different directions,and the influence of occlusion is reduced by the weighted sum of the histogram distance credibility.Finally,different depth map optimization methods are proposed for different scenarios.The experimental results show that the depth estimation results of the algorithm in the non-central sub-aperture images are better than the original algorithm in different scenarios.In order to analyze the influence of light field quality on depth estimation algorithm and rendering algorithm,this paper designs and implements a rendering system of compressed light field,which includes three parts: decoding of light field compressed data,rendering of light field and virtual reality display of rendered image.The light field rendering is done using depth information for interpolation.The system performs the depth estimation and rendering functions of different light field qualities.Subjective and objective quality analysis was performed on a rendered system of compressed light fields to analyze the effect of light field quality on depth estimation and rendering.The experimental results show that the proposed algorithm has certain robustness to the optical field quality,and maintains good performance under a certain range of light field compression rate.At the same time,under different compression ratios,the algorithm is always better than the original algorithm.