Subjective Quality Assessment of the Impact of Buffer Size in Fine-Grain Parallel Video Encoding

Abstract

Fine-Grain parallelism is essential for real-time video encoding performance. This usually implies setting a fixed buffer size for each encoded block. The choice of this parameter is critical for both performance and hardware cost. In this paper we analyze the impact of buffer size on image subjective quality, and its relation with other encoding parameters. We explore the consequences on visual quality, when minimizing buffer size to the point of causing the discard of quantized coefficients for highest frequencies. Finally, we propose some guidelines for the choice of buffer size, that has proven to be heavily dependent, in addition to other parameters, on the type of sequence being encoded. These guidelines are useful for the design of efficient realtime encoders, both hardware and software.