No reference objective quality metric for stereoscopic 3D video

Abstract

The stereoscopic three-dimensional (3D) video technologies have achieved significant success in providing enhanced immersive experience to consumers. However network delivery of 3D video content at good quality levels is challenging mostly due to the variable network conditions. In this context, efficient objective 3D video quality assessment is a critical aspect, in particular for video service providers who need to adjust the video delivery process to the network conditions in real-time. Current objective 3D video quality assessment methods are reference-based, requiring the availability of the original 3D video sequences, which is difficult to achieve in practice. Additionally, most of the existing 3D video quality metrics are developed for depth-enhanced 3D. This paper proposes the No reference objective Video Quality Metric (NVQM) for real-time 3D video quality assessment. NVQM considers the correlation between network packet loss and perceptual video quality for different bit-rate video sequences. NVQM is modeled based on the video quality model specified in ITU-T G.1070 and tuned according to results of extensive subjective tests. NVQM was developed for the evaluation of side-by-side stereoscopic 3D sequences, the most widely commercialized 3D video format. The performance of NVQM is studied by comparing against three state-of-the-art video quality objective models: structural similarity index (SSIM), video quality metric (VQM), and ITU-T G.1070. Results show that NVQM outperforms the existing objective metrics with up to 23% in terms of accuracy.