VVC

VVC is one of the core topics which MSP members are interested in. We have tried to invent various algorithms to increase the performance of the video codec. The following figure represents a block diagram of VVC encoder. We have many ideas to improve this codec and some of them have been submitted as a form of patent.

Multiview HEVC

The MV-HEVC standard was designed to encode two or more views, as shown in the following figure, where the pictures in neighboring views are used as the reference pictures to encode the current picture in other views. Several researches have been conducted to increase the performance of MV-HEVC. 

Because MV-HEVC supports multi-view video coding without using camera calibration information, there are some constraints in removing the redundancy between neighboring views that result in some inefficiency in inter-view coding between neighbor views. To overcome this problem, we focus on making an exact motion vector predictor between neighbor views.

Although the PMV derivation process of MV-HEVC derives PMVs by considering the types (i.e., STRP and LTRP) of reference pictures, the advanced motion vector prediction (AMVP) and merge processes of MV-HEVC still lack efficient PMVs. The reason is that the PMV derivation process of MV-HEVC does not discriminate between the types of inter-view reference pictures and temporal long-term reference pictures. This means that an inter-view reference picture is also considered an LTRP. Therefore, when MVPs of the temporal LTRP type are listed, inter-view disparity MVs can be included in the MVP list. Because the properties of the temporal LTRP and the inter-view reference picture are different from each other, this process may result in a reduction in coding performance. In this research, we propose an efficient algorithm to provide the exact MVPs by discriminating them. In order to increase the similarity between the current MV and the inter-view MVP, we exploit the geometric relationship between neighbor views.

HEVC

In the applications using UHD sequence, since the video quality fluctuation results in a major negative effect on the subjective evaluation, the quality control is one of the most important aspects related to HEVC. Thus, it is important to control the quality of the encoded frame efficiently in order to maintain consistent video qualities over the entire video sequence. In HEVC, because the quadtree-based CU structure is used and the temporal distance between the current and reference frames is variable according to the level of hierarchical B picture coding structure, the conventional schemes are not appropriate to control the quality and bit rate. In this research, we propose an efficient rate control algorithm to minimize the fluctuation of video quality in HEVC where the R-Q and D-Q models are derived considering the quadtree CU structure.

The following figures show PSNRs produced by rate control algorithms for hierarchical and low delay B picture structures, respectively, where the proposed rate control algorithm generates the smaller PSNR fluctuation over the entire sequence than other schemes.

The following figures show buffer fullness of the HEVC encoder when the rate control algorithms are used for hierarchical and low delay B picture structures, respectively. As we observe in these figures, the conventional algorithms generate many bits for some frames having high complex texture where overflows are occurred. Although some conventional methods control the generated bits properly in low delay B picture structures, underflows are occurred when they are used to encode B pictures in high levels of hierarchical B picture structures, because larger QPs are assigned to those frames without considering buffer status. These figures show the proposed rate control algorithm keeps buffer status under the upper limit of the buffer level, because the number of generated bits is accurately estimated by using the proposed R-Q model, and Rmin and Rmax are used to prevent the buffer from underflow and overflow.

Scalable Video Codec