CN-117478911-B - VVC multidirectional motion refinement optimization method based on Merge mode

Abstract

The invention discloses a VVC multidirectional motion refinement optimization method based on a Merge mode, which is used for searching a larger step length by only a small part of Coding Units (CUs) in the Merge process with motion vector difference in the VVC inter-frame prediction and fails to consider the multidirectional nature of motion in the actual process. The method comprises the steps of selecting first two pieces of MV information in an expanded Merge candidate list as initial MVs, expanding from four steps in eight directions, judging whether a current step index is larger than 1, if so, skipping a chroma block motion compensation and bidirectional optical flow prediction mode, and if not, skipping the chroma block motion compensation to obtain an optimal Merge mode, and then performing entropy coding. Compared with the VVC coding standard, the average bit rate saving of Y, cb and Cr is 0.57%, 0.62% and 0.25% respectively.

Inventors

• HE XIAOHAI
• TANG JIANXIANG
• HE HAIBO
• XIONG SHUHUA
• SUN WEIHENG
• TENG QIZHI
• GONG JIAN

Assignees

• 四川大学
• 成都西图科技有限公司

Dates

Publication Date

20260508

Application Date

20220719

Claims (3)

1. 1. A Merge mode-based VVC multidirectional motion refinement optimization method is characterized by comprising the following steps of: (1) Carrying out MMVD-PRO candidates, selecting the first two pieces of MV information in an expanded Merge candidate list as initial MVs, combining optimization step length selection and direction selection and expanding, namely reducing the original eight step lengths into first four smaller step lengths in the up-down left-right directions, wherein the four step lengths are kept unchanged, and respectively expanding the four step lengths in the up-down left-right directions in the inclined direction in the four step lengths in the four directions of the up-down left-right directions by half of the first four smaller step lengths; (2) And obtaining an optimal Merge mode based on motion compensation and rate distortion cost, and performing entropy coding on the finally selected optimal mode, wherein the searching step length and the searching direction are fully utilized to adjust entropy coding modes corresponding to different points of the coding modes, the searching step length index code length is reduced by half, the direction index code length is only increased by one bit, if the MMVD _PRO mode is selected, MMVD _PRO flag bits are coded, and otherwise, the flag bits of other modes are coded.
2. 2. The method for refining and optimizing the VVC multidirectional motion based on the Merge mode as claimed in claim 1, wherein the characteristic that step selection in the Merge mode is mainly concentrated on smaller step is utilized, the step search range is compressed, the multidirectional nature of motion in the actual process is fully considered, and the expression mode of motion characteristics in the oblique direction is increased.
3. 3. The Merge mode-based VVC multi-directional motion refinement optimization method according to claim 1, characterized in that the mode selection procedure is adaptively adjusted according to the rate-distortion cost criterion while taking into account the step size and direction selection manner of the two modes MMVD mode and MMVD _pro mode.

Description

VVC multidirectional motion refinement optimization method based on Merge mode Technical Field The invention relates to the technical problem of inter-frame prediction optimization in the field of video coding, in particular to a multi-functional video coding standard (VERSATILE VIDEO CODING, VVC) multi-directional motion refinement optimization method based on a Merge mode. Background With the increasing progress of communication technology and the rapid popularization of intelligent terminals, digital video is ubiquitous in life. The quality and resolution requirements of digital video are increasing, so that the emerging video coding technology has important significance in the present internet age. However, the conventional video compression coding techniques h.264/AVC and h.265/HEVC have failed to meet the market demands because of the transmission broadband and limited storage capacity of various terminal devices. Universal video coding (VERSATILE VIDEO CODING, VVC) has been released as a new generation video coding standard, 7 months 2020. Compared with the previous generation Video Coding standard, namely High Efficiency Video Coding (HEVC), the VVC introduces a plurality of new technologies to improve the compression efficiency, and the Coding performance is improved by about 50%. The overall coding framework of the VVC standard is generally consistent with HEVC, yet a block-based hybrid coding architecture is still in use. However, each link is improved or new technologies are added in the whole framework, and taking an inter-frame prediction module as an example, some important improvement measures are taken, namely, a motion vector prediction candidate based on historical information is newly added to the VVC, the length of an original Merge mode candidate list in HEVC is increased from 5 to 6, and the new frame is named as an extended Merge mode. Meanwhile, in order to improve the prediction accuracy, an inter-frame prediction signal and an intra-frame prediction signal are combined, an inter-frame combined prediction mode is newly added, and the information of the two prediction blocks is weighted through a weighting coefficient to obtain a final prediction value. The VVC is added with a Merge technology with motion vector difference, and the motion vector is represented by three parameters of a starting point, a step length and a motion direction so as to reduce the tiny deviation generated by the motion information and the motion of an actual coding block. In addition, the introduction of other new technologies also improves the prediction precision to a certain extent, such as symmetric motion vector difference coding, motion time domain motion vector prediction based on sub-blocks, bidirectional weighted prediction, bidirectional optical flow, triangular division mode, affine motion compensation prediction, self-adaptive motion vector precision and decoding end motion vector correction. Although the Merge mode can further compress data, certain errors are generated correspondingly, and there are still many deserves to be perfected, and many researches on the aspect are also performed in the current academia. In order to improve the prediction accuracy of the Merge mode, hong et al propose a new Merge mode decision algorithm, which uses a motion vector candidate algorithm derived from the decoding end to make a Merge mode decision based on absolute bi-directional prediction difference and a template matching criterion. Li et al propose a motion-assisted merge mode to obtain more accurate motion vectors by searching for small motions obtained in a particular search area. Shinobu Kudo et al have proposed an MV prediction method considering prediction continuity in order to improve coding efficiency by setting a MV prediction candidate list as a block coded in an intra prediction mode, considering that motion compensation efficiency in a Merge mode is reduced if a reference block is coded in the intra prediction mode. Fan et al found the shortcomings of the existing intra prediction in HEVC, interpreted the coding gain reasons for boundary filtering, and established a filtering method on the boundary using a generalized gaussian distribution function, and extended the proposed boundary filter to the inter prediction Merge process. Park et al propose an efficient method of constructing a Merge mode candidate list that reduces the use of zero motion vectors as candidate lists by generating an alternative Merge candidate list. Disclosure of Invention In the VVC inter prediction process, in the Merge mode prediction process, only a small part of Coding Units (CUs) with motion vector differences select a larger step size search, and the larger step size needs to be represented by a larger number of bits, resulting in a reduction in prediction performance. Aiming at the problem that multidirectional motion exists in the actual process is not fully considered in inter-frame prediction of the VVC, the