Quality Assessment And Coding Optimization For Omnidirectional Video

Recently,virtual reality(VR)has attracted much attention and VR applications have been a topic of increasing interest.Omnidirectional video,one of the most primary media contents of VR,usually expects high resolution and high frame rate to guarantee the immersive viewing experience.And omnidirectional video normally needs to be projected to a 2-D plane before video processing,adapting to the input of existing video processing technologies,which may introduce deformations and information redundancy.Thus,correlated video processing technologies for omnidirectional video are essential in the future development of VR.Specialized and superior objective video quality assessment(VQA)in the spherical domain for omnidirectional video can help measure the quality of video processing algorithms,and high efficiency coding optimization can reduce the resource consumption of omnidirectional video storage and transmission,which are two critical issues in the development of omnidirectional video processing.Thus,our work and contributions are mainly divided into the following two parts.First,we propose a new spherical structural similarity measurement(S-SSIM)for omnidirectional VQA.We propose the method of calculating structural similarity of each pixel in the spherical domain.Then we analyze the method of calculating structural similarity of the whole image in the spherical domain to remove the influence of projection and obtain the final structural similarity measurement.We utilize two subjective omnidirectional VQA databases with completely different test sequences to verify the performance of S-SSIM.Meanwhile,we also verify the performance of the traditional objective VQA metrics and the state-of-the-art objective omnidirectional video quality evaluation metrics.Through the analysis and comparison of experimental results,we can prove that our proposed S-SSIM algorithm is the most consistent with the subjective scoring results and has the best performance.Second,we apply the proposed S-SSIM to improve the rate distortion optimization(RDO)processing in video coding.Considering the distortion index in the existing ratedistortion optimization are very simple and not consistent with the subjective evaluation results,we utilize the proposed S-SSIM metric to replace distortion measurement in rate-distortion cost computation to select the best block partition mode of each coding unit(CU).To establish a new rate distortion model,we retrain the new relationship between Lagrange multiplier and quantization parameter and propose the new RDO method.Then we apply the proposed new RDO method to the block partition decision process in the existing video coding framework to select the best block partition mode for each CU.Based on the experimental results,compared with the traditional coding algorithm,our proposed coding optimization algorithm can improve coding efficiency.We also provide visual examples to intuitively show the quality of different omnidirectional videos compressed by different algorithms,and the visual examples show that our proposed coding optimization algorithm can improve the subjective quality of the compressed video.