Evaluation Of Light Field Quality Based On Images From New Perspectives

In recent years,with the rapid development of 3D video services,the light field,as a new three-dimensional scene representation method,has received increasing attention from people.The light field can simultaneously provide three kinds of visual information: mobile parallax,binocular disparity,and selective focus.With image-based rendering technology,light field imaging can generate multi-viewpoint,multi-focal-plane,multi-aperture new perspective images,and will image The result has been raised from the traditional 2D plane image to the 4D light field image.However,the image-based rendering technology can only generate the image quality that meets the requirements under the premise of high light field sampling density.Therefore,it is very important to design a rendering algorithm that can also improve the rendering quality when the sampling density is low.Compared with the 3DOF feature of traditional 3D video services,light field imaging enables 6DOF free-view video services,providing a more realistic and comprehensive visual experience.The quality of the light field directly affects the user's final visual experience.The evaluation of the quality of the light field is critical to the performance of the light field acquisition,compression,and rendering algorithms.This thesis first proposes a light field interpolation algorithm based on depth information for the light field rendering algorithm,and then designs and implements a subjective evaluation system for light field quality based on VR equipment,and deeply analyzes the factors affecting the light field quality and their mutual relations.In order to improve the imaging quality of the light field,a rendering algorithm based on depth information is proposed in this thesis.First we need to get the depth information of the light field source.Based on the two major light field depth prediction algorithms,a complete depth map of the light field data source is generated,and the performance differences of different depth prediction algorithms are analyzed.Then based on the light field data source with depth information,the rendering algorithm is designed and implemented.Different from the conventional IBR technique of distance-prioritized light selection criterion,the depth-based interpolation filter kernel function selects the sampling light according to the depth difference when constructing.Finally,through the analysis of the rendering results,we can see that the PSNR of the rendering algorithm presented in this thesis has increased by at least 5 dB compared to the IBR rendering algorithm,and the image quality of the new perspective has been greatly improved.For the quality evaluation of the light field,this thesis proposes and builds a light field experience quality evaluation system based on VR equipment.The system fully evaluates the subjective quality of the light field in the form of stereo vision and a completely free viewpoint,which improves the accuracy of subjective evaluation results.Afterwards,by designing subjective experiments,the relationship between depth information,scene content and subjective experiment MOS was studied.Experimental results show that the performance of rendering algorithm based on depth information is obviously better than that of IBR without depth information.The difference of depth information also affects the rendering algorithm proposed in this thesis.The more accurate the depth information is,the higher the rendering quality is;The view image quality is closely related to the scene content.The scene content reflected by the scene texture complexity and the scene depth range shows that the higher the texture complexity and the greater the depth range,the worse the imaging quality.For the objective evaluation of light field quality,an objective evaluation model considering the quality of the depth map and the scene content is proposed.By constructing the relationship between the quality of the light field new perspective image and the objective influencing factor,the conclusion that is consistent with the subjective experiment is obtained,so that the quality of the light field is more fully predicted.