One-pass mode decision for low-complexity and high-efficiency encoding of quality scalable video

Abstract

This paper presents a novel one-pass mode decision algorithm for encoding multiple quality layers, following the coarse grain scalability coding structure in the H.264/SVC standard. In our solution, motion estimation (ME) is carried out only once at the base layer using the reconstructed picture from the highest enhancement layer. The same motion vectors are used for all layers to not only avoid multiple ME processes at different layers, but also save the overhead bits. In addition, early SKIP/DIRECT mode decision is introduced to further boost the encoding speed. The encoder produces fully compliant SVC bit streams. Although the method is applicable to both coarse grain scalability (CGS) and medium grain scalability (MGS), we have examined its performance over CGS only. We demonstrate that more than 2x speedup for three-layer coding against the conventional H.264/SVC encoding using the reference software over 7 test sequences. Significantly, this complexity saving is achieved simultaneously with increase in the coding efficiency! Although the base layer requires slightly higher bit rate (2.5% in terms of the BD-Rate), the enhancement layers enjoy lower rates (5.7% and 2.2% reduction for the total of two and three layers, respectively), on average of 7 test sequences.