Viewport-Based Omnidirectional Video Quality Assessment: Database, Modeling and Inference

水虾虾虾

已于 2022-03-25 17:52:27 修改

阅读量165

点赞数

文章标签：视频编解码

于 2022-03-25 17:51:34 首次发布

原文链接：https://ieeexplore.ieee.org/document/9349097

版权

一、提供数据库

1.VOD-VQA Database

18 ODVs --> encode with different STARs --> 774PVSs 用于主观评分

1 ODV 用于训练

2.主观质量研究

MOS: the mean opinion scores

MOS(s, t, q)

二、建立分析模型

设计度量标准，预测主观质量

$Q_{\mathrm{STAR}}^{\mathrm{VP}}$ : viewport-based quality index

1.建立模型

separable impacfts s, t, q respectively

$\begin{aligned} \operatorname{NQS}(s, q, t) &=\frac{\operatorname{MOS}(s, q, t)}{\operatorname{MOS}\left(s_{\max }, q, t\right)} \\ \operatorname{NQQ}(s, q, t) &=\frac{\operatorname{MOS}(s, q, t)}{\operatorname{MOS}\left(s, q_{\min }, t\right)} \\ \operatorname{NQT}(s, q, t) &=\frac{\operatorname{MOS}(s, q, t)}{\operatorname{MOS}\left(s, q, t_{\max }\right)} \end{aligned}$

the joint impacts

$\begin{aligned} Q_{\mathrm{STAR}}^{\mathrm{VP}} &=Q(s, t, q)=\frac{\operatorname{MOS}(s, t, q)}{\operatorname{MOS}\left(s_{\max }, t_{\max }, q_{\min }\right)} \\ &=\hat{Q}_{\mathrm{NQQ}}\left(q ; t_{\max }, s_{\max }\right) \hat{Q}_{\mathrm{NQS}}\left(s ; t_{\max }, q\right) \hat{Q}_{\mathrm{NQT}}(t ; s, q) \end{aligned}$

$\hat{Q}_{\mathrm{NQQ}}\left(q ; t_{\max }, s_{\max }\right)=\hat{Q}_{\mathrm{NQQ}}(q)=\frac{1-e^{-\alpha_{q} \cdot \hat{q}^{\beta q}}}{1-e^{-\alpha_{q}}}$

$\hat{Q}_{\mathrm{NQS}}\left(s ; t_{\max }, q\right)=\hat{Q}_{\mathrm{NQS}}(s ; q)=\frac{1-e^{-\alpha_{s} \cdot \hat{s}^{\beta s}}}{1-e^{-\alpha_{s}}}$

$\hat{Q}_{\mathrm{NQT}}(t ; s, q)=\frac{1-e^{-\alpha_{t} \cdot \hat{t} \beta_{t}}}{1-e^{-\alpha_{t}}}$