Unified distributed source coding frames for interactive multiview video streaming

摘要

Because of differential coding used in standard video compression algorithms to exploit temporal correlation in adjacent frames for coding gain a frame lost in network will cause error propagation in subsequent frames at the decoder Previously proposed distributed source coding (DSC) frames can be periodically inserted to halt this error propagation by overcoming the uncertainty at encoder of which frames will be correctly received at decoder without resorting to large intra-coded I-frames In the case of interactive multiview video streaming (IMVS) where a user watches one of M available captured views at a time but can periodically select and switch to a neighboring view the encoder must encode multiview video to enable this view-switching interactivity without knowing the exact view trajectories taken by viewers at stream time In this paper we propose a unified DSC frame construction for IMVS so that the encoder can overcome both types of uncertainty in a coding-efficient manner; ie halt error propagation in differentially coded multiview video and facilitate periodic interactive view-switching at the same time Having the additional unified DSC frames we design a multiview frame structure to maximize the expected number of correctly decoded frames at decoder for a given bandwidth constraint We develop a fast algorithm to find locally optimal structure parameters and packetization and packet reordering strategies for transmission Experimental results show that our optimized frame structures using unified DSC frames outperform naïve structures using I- and P-frames only by up to 49% in fraction of correctly decoded frames under typical network condition