Joint Collaborative Team on Video Coding (JCT-VC)
of ITU-T SG16 WP3 and ISO/IEC JTC1/SC29/WG11
11th Meeting: Shanghai, CN, 10–19 Oct. 2012
Document: JCTVC-K0103
Title: Rate control by R-lambda model for HEVC
Status: Input Document to JCT-VC
Purpose: Proposal
Author(s) or
Contact(s):
Bin Li
Houqiang Li
Li Li
Jinlei Zhang
Email: yhlibin@mail.ustc.edu.cn
lihq@uste.edu.cn
Source: University of Science and Technology of China
_____________________________
Abstract
This contribution proposes R-λ model based rate control for HEVC. The proposed rate control algorithm
is implemented on HM-8.0. Compared with the existing rate control algorithm in HM-8.0, when targeting
the bitrate of HM-8.0 default anchor (without rate control), the proposed method obtains 0.42 dB ~ 1.12
dB Y PSNR gain for different cases. The bitrate errors (the difference of the target bitrate and the actual
bitrate) of the proposed method, 0.09%~0.22% for different cases, are also much smaller than those of the
existing rate control algorithm in HM-8.0, which are 0.16%~1.09% for different cases.
1 Introduction
The existing rate control algorithm in HM-8.0 applies a unified RQ model first proposed in JCTVC-
H0213 [1]. But according to previous investigation in JCTVC-I0426 [2], λ is more important than QP
value on determining the bitrate. Thus, this contribution proposes a new rate control algorithm for HEVC
via R-λ model. The R-λ model will be described in Section 2, and the proposed rate control algorithm via
R-λ model will be described in Section 3. Section 4 provides detailed experimental results and analysis.
2 R-λ model
This contribution will build the relationship between R and λ. As we know, several types of R-D model
have been proposed to characterize the relationship between R and D. One typical R-D model is
Exponential function [4],
(1)
where C and K are model parameters related to the characteristic of the source. Some researchers also
model the R-D relationship by Hyperbolic function [5][6],
(2)
where C and K are also model parameters related to the characteristic of the source, but are different from
the C and K in (1).
This contribution performs some experiments to compare these two models based on HEVC. We use the
simple IBBB coding structure with only one reference picture and flat QP in HM-7.0, and then fit the R-D
curve according to the two different models, as shown in Fig. 1. R is expressed in terms of bpp, and D is
expressed in terms of MSE (Mean Square Error) of Luma component. They are calculated by (3) and (4)
Page: 1 Date Saved: 2012-09-23