Research On Quality Assessment Of 3D Point Cloud Without Reference Based On Bitstrea

With the advancement of science and technology and the development of technology.The application of computer graphics driven by 3D point cloud in the fields of reverse engineering,digital city,cultural relics protection,intelligent robot,unmanned driving,human-computer interaction and other fields has gradually appeared in people’s sight.The demand for bandwidth-constrained 3D point cloud applications is also rapidly increasing,and it is becoming more and more important to accurately determine the quality of compressed point clouds and evaluate the quality to optimize the user experience.In order to achieve effective monitoring of 3D point cloud quality.In this paper,we deeply study the quality assessment methods of 3D point clouds,and propose two bitstream-based noreference 3D point cloud quality assessment methods: No-reference Quality Assessment of 3D Point Clouds Based on v-SVR and Bitstream-based Perceptual Quality Assessment of Compressed 3D Point Clouds.The major contributions of this dissertation are summarized as follows:(1)In order to obtain more accurate 3D point cloud quality,a reference-free 3D point cloud quality evaluation model based on v-SVR is proposed.This model is suitable for point clouds compressed by G-PCC(Octree+LT).Firstly,the relationship between encoding parameters from the distorted point cloud and the subjective quality is analysed.Then the impact of the encoding settings on the subjective quality can be obtained.Secondly,the impact of content characteristic on the subjective quality can be determined under the same encoding settings.Then two models to estimate the geometric and textural characteristic factors respectively is proposed,both of which can characterize the point cloud content.Finally,we use quantization parameters,location quantization scales,texture feature factors,and geometric feature factors as the input parameters of vSVR and the subjective quality scores as the output parameters.As a result,we get a noreference quality assessment model of 3D point clouds which can reflect human visual characteristics.(2)In this thesis,we describe a bitstream-based no reference model for perceptual quality assessment of point clouds without resorting to fully decoding.We firstly analyze the nonlinear relationship between geometric lossless Mean Opinion Score(MOS)and the textural quantization parameter.We then use the textural quantization parameter and the textural bitrate to estimate the textural complexity of a point cloud according to the rate-distortion theory.Based on the human visual masking effect,we propose a geometric lossless textural distortion assessment model only utilizing the textural quantization parameter and the textural bitrate.We finally propose a Bitstream-Based No Reference model of Point Cloud Quality Assessment(BBNR-PCQA)after analyzing the geometric distortion effect caused by Trisoup.This thesis proposes two kinds of no reference model of point cloud quality assessment,which have the characteristics of low computational complexity and high performance and can be applied in most scenarios and applications.Considering the spatial masking effect produced by human vision system,a parameter which can explain the content complexity of point cloud is proposed,and the advantages of the proposed model in terms of performance and time complexity are proved by experiments.