A no reference depth perception assessment metric for 3D video

Nur Yılmaz G.

MULTIMEDIA TOOLS AND APPLICATIONS, vol.74, no.17, pp.6937-6950, 2015 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 74 Issue: 17
  • Publication Date: 2015
  • Doi Number: 10.1007/s11042-014-1945-y
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED)
  • Page Numbers: pp.6937-6950
  • Keywords: 3D video, Aerial perspective, Binocular parallax, Depth perception, Gaussian Mixture Model (GMM), Expectation Maximization (EM), QoE, Lateral motion, REFERENCE QUALITY ASSESSMENT, H.264/AVC ENCODED VIDEO, BINOCULAR VISION, PSNR, DISTORTION, MOTION, DOMAIN
  • TED University Affiliated: No


Recent technological breakthroughs in 3-Dimensional (3D) video capture, display, coding, transmission, rendering, etc. have led the advances of 3D multimedia applications into the consumer market. However, the effect of these technologies on 3D video Quality of Experience (QoE) has not been thoroughly investigated to speed up the wide-spread proliferation of the 3D video applications in this market. Quality and depth perception assessment of 3D video from the view of end users reflects the most important aspect of 3D video QoE. Therefore, evaluating quality and depth perception of 3D video should be given the utmost attention. Currently, the depth perception assessment of 3D video can only be achieved using time consuming and rigorous subjective assessments due to the lack of reliable and efficient objective metrics. Assessing the depth perception using Full-Reference (FR)/Reduced Reference (RR) objective metrics is not efficient for on the fly 3D video applications due to the requirement of original video/extracted information at the receiver side. Thus, a No Reference (NR) metric, which does not need any original video related information at the receiver side to predict the depth perception, is proposed in this paper. Three important cues (i.e., binocular parallax, lateral motion, and aerial perspective) for Human Visual System (HVS) to perceive the depth of a 3D video are utilized to develop the NR metric. Experimental results devised using the proposed metric prove the effectiveness of it to predict the depth perception.