Abstract
Conference Title: 2016 IEEE International Conference on Multimedia & Expo Workshops (ICMEW) Conference Start Date: 2016, July 11 Conference End Date: 2016, July 15 Conference Location: Seattle, WA, USA In interactive multiview video streaming (IMVS), a client periodically requests switches to neighboring views for uninterrupted temporal video playback from a server storing a large number of pre-encoded views. The technical challenge is that the navigation path taken by a client is not known at encoding time, and thus it is difficult to employ differential coding to lower code rate without knowing exactly what frames are available at the client buffer as predictors. In a previous work, a new frame type called merge frame was designed to efficiently merge different side information (SI) frames Sn from different possible decoding paths into a unique construction M, so that the following frame(s) in time can be differentially coded using M as predictor without coding drift. In this paper, we design new coding structures using two variants of merge frame for different view-switching probabilities and desired rate-distortion (RD) tradeoff points. Experimental results show that our proposed frame structure designs outperform view-switching mechanisms in the literature, such as SP-frames in H.264, in RD performance.