Rate-Distortion Models for FGS-encoded Video Sequences

Fine granularity scalability (FGS) coding enables finer bitrate scalability and better error resiliency than traditional multi-layer coding, because it allows truncating a bitstream at arbitrary bits. This flexibility enables streaming applications a more space to optimize quality (i.e., minimize distortion) at a given channel bandwidth. This optimization relies on accurate estimation of the rate-distortion (R-D) characteristics of video sequences. In this paper, we analyze and compare the performance of the R-D models proposed in the literature for FGS coding systems. We analyze the models by following their mathematical derivations and scrutinizing their assumptions. We perform the comparison by implementing the models and conducting an extensive experimental study using a large set of video sequences with diverse image and motion complexities. The results of our experiments provide guidelines for choosing the appropriate R-D model for rate-distortion optimized streaming applications.