Estimation of the Optimum Depth Quantization Parameter for a Given Bitrate of Multiview Video Plus Depth in 3D-HEVC Coding

Abstract

One of the most popular data formats to represent natural immersive visual data is the MVD (Multiview Video plus Depth) format. The representation of a three-dimensional scene requires a huge amount of data in the form of a dense set of views accompanied by high-quality depth maps. All of this data need to be transmitted over the network to a viewer. Therefore, a question arises, how to allocate the bitrate between views and depth maps to obtain the maximal quality for a given bit budget. In the paper, a simple model for optimum bitrate allocation between color and depth data in 3D-HEVC coding is proposed. The provided model for quantization parameters allows better bitrate division between color and depth data, which leads to a significant (23-35%) bitrate reduction of the total bitrate of the multiview stream. At the same time, it preserves the same quality of synthesized virtual views in comparison to the common test condition (CTC) recommendation, which is considered as a wellestablished reference encoder configuration.

Universal Modeling of Monoscopic and Multiview Video Codecs with Applications to Encoder Control

“…Similarly, the quantization parameters QD for depth may be the same for all the corresponding depth maps. Experimental data analysis leads to a conclusion that some (QP, QD) pairs correspond to the curve of the maximum PSNR [36], e.g., as in Fig. 3.…”