SUBJECTIVE QUALITY ASSESSMENT OF MOBILE 3D VIDEOS Zhiwei Fan, Tingting Jiang National Engineering Laboratory For Video Technology, School of EECS, Peking University ABSTRACT With the development of mobile network and 3D display technology, watching 3D videos on portable devices through mobile network is to be a trend. In this paper, we study the influence of Quantization Parameters (QP), resolution and packet loss on the QoE of mobile 3D videos. Side-by-side formatted 3D video sequences of different QP and resolution, transmitted through networks of different packet loss ratios (PLR) are produced, and two subjective assessment experi- ments have been made on two portable devices. Each video sequence is tested according to four measures, including clearness, depth perception, comfort level and overall quality. The results show that QP and resolution have a greater impact on mobile 3D videos’ clearness than depth perception and that the influence of packet loss is not as stable as the other factors. Index Terms— mobile, 3D, QoE, transmission, subjec- tive 1. INTRODUCTION With the development of mobile network, watching videos on mobile or portable devices through mobile network is be- coming more and more popular. In the meanwhile, with the development of 3D display technology, 3D videos are wide- ly appreciated, and people want to watch 3D videos more than in the cinema. As some commercially available portable devices currently support auto-stereoscopic display function, watching mobile 3D videos on portable devices is to be the trend. [1] tried to examine users’ needs and requirements for mobile 3D television and videos in a psychological way, and users’ desires for having 3D content everywhere could be proved. In that case, it is important to find a way to offer mobile 3D videos with high QoE, and getting knowledge of the QoE of mobile 3D videos is a first step. There have been a number of databases related to 3D videos or mobile videos, while a database specifically de- signed for both 3D and mobile videos is quite rare. [2–4] each described databases of 3D videos that encompassed a range of environments, lighting conditions, textures, motion, coding conditions, capture parameters, etc. However, none This work is partially supported by NSFC (61390514, 61103087, 91120004, 61210005) of them was designed for mobile displays and the impair- ments due to transmission were not introduced. [5] presented a database specifically designed for 2D mobile TV quality as- sessment, and [6] presented a subjective assessment database of videos with lost slices due to network impairments. But they were both designed for 2D videos. Therefore a new database of both mobile and 3D videos is really in need. There are a series of artifacts that influence the QoE of mobile 3D videos, caused in different stages of 3D video pro- duction and delivery, including content creation, conversion to the desired format, coding/encoding, transmission, and vi- sualization on 3D display [7]. And the artifacts that are most pronounced on portable displays have been analyzed in [8]. In the process of content creation, unnatural disparity be- tween the views in the stereo-pair is the most common and an- noying artifact for 3D videos, which leads to bad performance in depth perception. Improper camera geometry and position often cause this problem. [2] offered a database, where se- quences of each scene were captured with different camera distances, and a subjective test had been made on it. Unfortu- nately, the test was made on a computer monitor, and no such experiments on portable displays have been found. Besides, changing the resolution of sequences or the size of displays can introduce unnatural disparity as well [8]. It means that a 3D video with excellent QoE played on large displays like TV screens, may perform badly, especially in depth perception, on small displays like phones. In that case, the impact of dis- plays size and video resolution on QoE of mobile 3D videos, especially on depth perception, is an important research is- sue, as many videos played on mobile devices are originally created for larger displays, for example 3D movies. In the process of format conversion and coding/encoding, more artifacts, compression artifacts for instance, are intro- duced. [9] compared the performance of two video formats (frame-sequential and side-by-side) and two codecs (JMVC and JM) for 3D videos. [10, 11] evaluated four different cod- ing methods and two codec profiles for mobile 3D-TV, finding that Multi View Coding and Video + Depth Coding performed better than the others, and that the high profile could provide the same quality as low profile at a lower bit rate. Compres- sion artifacts are the main artifacts in the process of coding, on which various researches have been done. [12,13] explored the influence of varying depth and compression artifacts on the QoE of mobile 3D video content, showing that compres-