Single-Pass Dependent Bit Allocation in Temporal Scalability Video Coding
Jiaying Liu,Yongjin Cho,Zongming Guo
DOI: https://doi.org/10.1109/DCC.2013.87
2013-01-01
Abstract:Summary form only given. In the scalable video coding, we refer to a group-of-pictures (GOP) structure that is composed of hierarchically aligned B-pictures. It employs generalized B-pictures that can be used as a reference to following inter-coded frames. Although it introduces a structural encoding delay of one GOP size, it provides much higher coding efficiency than the conventional GOP structures [2]. Moreover, due to its natural capability of providing the temporal scalability, it is employed as a GOP structure of H.264/SVC [3]. Because of the complex inter-layer dependence of hierarchical B-pictures, the development of an efficient and effective bit allocation algorithm for H.264/SVC is a challenging task. There are several bit allocation algorithms that considered the inter-layer dependence in the literature before. Schwarz et al. proposed the QP cascading scheme that applies a fixed quantization parameter (QP) difference between adjacent temporal layers. Liu et al. introduced constant weights to temporal layers in their H.264/SVC rate control algorithm. Although these algorithms achieve superior coding efficiency, they are limited in two aspects. First, the inter-layer dependence is heuristically addressed. Second, the input video characteristics are not taken into account. For these reasons, the optimality of these bit allocation algorithms cannot be guaranteed. We propose a single-pass dependent bit allocation algorithm for scalable video coding with hierarchical B-pictures in this work. It is generally perceived that dependent bit allocation algorithms cannot be practically employed due to their extremely high complexity requirement. To develop a practical single-pass bit allocation algorithm, we use the number of skipped blocks and the ratio of the mean absolute difference (MAD) as features to measure the inter-layer signal dependence of input video signals. The proposed algorithm performs bit allocation at the target bit rate with two mechanisms: 1) th- GOP based rate control and 2) adaptive temporal layer QP decision. The superior performance of the proposed algorithm is demonstrated by experimental results, which is benchmarked by two other single-pass bit allocation algorithms in the literature. The rate and the PSNR coding performance of the proposed scheme and two benchmarks at various target bit rates for GOP-4 and GOP-8, respectively. We see that the proposed rate control algorithm achieves about 0.2-0.3dB improvement in coding efficiency as compared to JSVM. Furthermore, the proposed rate control algorithm outperforms Liu's Algorithm by a significant margin.