A rate control algorithm for scalable video coding

TL;DR: This thesis proposes a rate control algorithm for H.264/scalable video coding (SVC) specially designed for real-time variable bit rate (VBR) applications with buffer constraints that aims to simultaneously control the buffer levels for overflow and underflow prevention, while maximizing the reconstructed video quality of the corresponding sub-streams.

Abstract: This thesis proposes a rate control (RC) algorithm for H.264/scalable video coding (SVC) specially designed for real-time variable bit rate (VBR) applications with buffer constraints. The VBR controller assumes that consecutive pictures within the same scene often exhibit similar degrees of complexity, and aims to prevent unnecessary quantization parameter (QP) fluctuations by allowing for just an incremental variation of QP with respect to that of the previous picture. In order to adapt this idea to H.264/SVC, a rate controller is located at each dependency layer (spatial or coarse grain scalability) so that each rate controller is responsible for determining the proper QP increment. Actually, one of the main contributions of the thesis is a QP increment regression model that is based on Gaussian processes. This model has been derived from some observations drawn from a discrete set of representative encoding states. Two real-time application scenarios were simulated to assess the performance of the VBR controller with respect to two well-known RC methods. The experimental results show that our proposal achieves an excellent performance in terms of quality consistency, buffer control, adjustment to the target bit rate, and computational complexity. Moreover, unlike typical RC algorithms for SVC that only satisfy the hypothetical reference decoder (HRD) constraints for the highest temporal resolution sub-stream of each dependency layer, the proposed VBR controller also delivers HRD-compliant sub-streams with lower temporal resolutions.To this end, a novel approach that uses a set of buffers (one per temporal resolution sub-stream) within a dependency layer has been built on top of the RC algorithm.The proposed approach aims to simultaneously control the buffer levels for overflow and underflow prevention, while maximizing the reconstructed video quality of the corresponding sub-streams. This in-layer multibuffer framework for rate-controlled SVC does not require additional dependency layers to deliver different HRD-compliant temporal resolutions for a given video source, thus improving the coding efficiency when compared to typical SVC encoder configurations since, for the same target bit rate, less layers are encoded. Resumen extendido en castellano