Perceptual Quality-of-Experience of Stereoscopic 3D Images and Videos

Abstract

With the fast development of 3D acquisition, communication, processing and display technologies, automatic quality assessment of 3D images and videos has become ever important. Nevertheless, recent progress on 3D image quality assessment (IQA) and video quality assessment (VQA) remains limited. The purpose of this research is to investigate various aspects of human visual quality-of-experience (QoE) when viewing stereoscopic 3D images/videos and to develop objective quality assessment models that automatically predict visual QoE of 3D images/videos.

Firstly, we create a new subjective 3D-IQA database that has two features that are lacking in the literature, i.e., the inclusion of both 2D and 3D images, and the inclusion of mixed distortion types. We observe strong distortion type dependent bias when using the direct average of 2D image quality to predict 3D image quality. We propose a binocular rivalry inspired multi-scale model to predict the quality of stereoscopic images and the results show that the proposed model eliminates the prediction bias, leading to significantly improved quality predictions.

Second, we carry out two subjective studies on depth perception of stereoscopic 3D images. The first one follows a traditional framework where subjects are asked to rate depth quality directly on distorted stereopairs. The second one uses a novel approach, where the stimuli are synthesized independent of the background image content and the subjects are asked to identify depth changes and label the polarities of depth. Our analysis shows that the second approach is much more effective at singling out the contributions of stereo cues in depth perception. We initialize the notion of depth perception difficulty index (DPDI) and propose a novel computational model for DPDI prediction. The results show that the proposed model leads to highly promising DPDI prediction performance.

Thirdly, we carry out subjective 3D-VQA experiments on two databases that contain various asymmetrically compressed stereoscopic 3D videos. We then compare different mixed-distortions asymmetric stereoscopic video coding schemes with symmetric coding methods and verify their potential coding gains. We propose a model to account for the prediction bias from using direct averaging of 2D video quality to predict 3D video quality. The results show that the proposed model leads to significantly improved quality predictions and can help us predict the coding gain of mixed-distortions asymmetric video compression.

Fourthly, we investigate the problem of objective quality assessment of Multi-view-plus-depth (MVD) images, with a main focus on the pre- depth-image-based-rendering (pre-DIBR) case. We find that existing IQA methods are difficult to be employed as a guiding criterion in the optimization of MVD video coding and transmission systems when applied post-DIBR. We propose a novel pre-DIBR method based on information content weighting of both texture and depth images, which demonstrates competitive performance against state-of-the-art IQA models applied post-DIBR.