阅读说明：本技术 360度视频帧内预测模式决策方法 (360-degree video intra-frame prediction mode decision method ) 是由 艾达 王易檀 胥策 艾宇 张晓阳 于 2021-06-30 设计创作，主要内容包括：一种360度视频帧内预测模式决策方法,由构建大尺寸预测单元模式决策方法、构建小尺寸预测单元模式决策方法步骤组成。构建大尺寸预测单元模式决策方法包括构建第一阶段的粗略模式决策方法、构建第二阶段的粗略模式决策方法、用于率失真优化的模式确定、确定最优模式步骤,构建小尺寸预测单元模式决策方法包括确定粗略模式决策的模式集合、确定用于率失真优化的模式步骤、确定最优模式步骤。预测单元PU尺寸分为两种大小,并且对不同尺寸PU根据其特点进行模式的预筛选,有效地减少了将粗略模式决策与率失真优化RDO过程所遍历的模式数目,实现了预测单元PU最优预测模式的快速决策。(A360-degree video intra-frame prediction mode decision method comprises the steps of constructing a large-size prediction unit mode decision method and constructing a small-size prediction unit mode decision method. The method for constructing the large-size prediction unit mode decision comprises the steps of constructing a rough mode decision method in a first stage, constructing a rough mode decision method in a second stage, determining a mode for rate distortion optimization and determining an optimal mode, and the method for constructing the small-size prediction unit mode decision comprises the steps of determining a mode set of the rough mode decision, determining a mode for rate distortion optimization and determining an optimal mode. The sizes of the prediction units PU are divided into two sizes, and the mode pre-screening is carried out on the PUs with different sizes according to the characteristics of the PUs, so that the number of modes traversed by a rough mode decision and a Rate Distortion Optimization (RDO) process is effectively reduced, and the quick decision of the optimal prediction mode of the prediction units PU is realized.)

1. A360-degree video intra prediction mode decision method is characterized by comprising the following steps:

(1) decision-making method for constructing large-size prediction unit mode

The prediction unit PU has a size of 64 × 64, 32 × 32, and a two-stage coarse mode decision method is adopted for the current prediction unit PU:

1) coarse mode decision method for constructing first stage

The 8 prediction modes 0, 1, 2, 10, 18, 26, 30, 34 are selected for the coarse mode decision in the first stage, and the absolute values of the hadamard transforms and SATD for the 8 prediction modes are determined as follows:

h is a normalized NxN-order Hadamard matrix, N is the size of the matrix, N is 4 or 8, X is a prediction residual matrix corresponding to a prediction mode and has the same size as H, p represents a row of HXH, and q represents a column of HXH;

two minimum patterns are selected from the absolute values and SATD of the 8 prediction modes, which are respectively called as first patterns FM₁Second mode SM₁；

2) Coarse mode decision method for constructing second stage

The mode set M for the second stage coarse mode decision RMD is constructed as follows:

carrying out rough mode decision of a second stage by using a mode set M;

3) determining the absolute value and SATD of 4 Hadamard transform according to formula (1) in the candidate mode set M, and selecting two modes FM with the minimum absolute value and SATD of Hadamard transform₁And SM₁Constructing a set of modes M for rate-distortion optimized RDO using the following equation_R：

4) Determining an optimal pattern

Adding the most probable mode MPM to the mode set M_RIn the method, a rate-distortion cost RDcost is determined according to equation (2), and the mode with the minimum cost value is found to be the optimal mode:

RDcost＝S_mode+λ×B'_mode (2)

wherein Smode represents the sum of squares of residuals between an original pixel block and a reconstructed block of a prediction unit PU, λ is a coding coefficient, and B' mode represents the number of bits consumed by intra-frame coding of the prediction unit PU;

(2) decision-making method for constructing small-size prediction unit mode

The size of the prediction unit PU is 16 multiplied by 16, 8 multiplied by 8 and 4 multiplied by 4, the texture direction of the prediction unit PU is preliminarily analyzed by a pixel absolute deviation method to obtain a matched angle mode, the angle mode adjacent to the angle mode, the mode 0 and the mode 1 form a mode for rough mode decision, the absolute value of Hadamard transform of each mode and SATD are compared to obtain an adaptive mode screening result, the adaptive mode screening result and the most probable mode MPM form a mode for rate distortion optimization, a rate distortion cost value RDcost is determined according to the formula (2) by a rate distortion optimization method, and the optimal mode is selected.

2. The 360 degree video intra prediction mode decision method of claim 1, wherein in (2), the method for constructing the mode for coarse mode decision is:

the mode set for determining the coarse mode decision is as follows:

determining pixel absolute deviation values ADP of all pixel points in the current prediction unit PU in 8 texture directions by using a pixel absolute deviation method, wherein the pixel absolute deviation values ADP of the 8 prediction directions respectively use horizontal absolute deviation values A_HorAbsolute vertical deviation value A_VerAbsolute deviation value of lower left diagonal A_DDLAbsolute deviation value of upper left diagonal A_DULAbsolute deviation value A of horizontal deviation from the upper_{H_U}Horizontal deviation from the lower absolute deviation value A_{H_D}Absolute deviation value A of vertical deviation from left_{V_L}Absolute deviation value A of vertical deviation to right_{V_R}Represents;

(1) the horizontal absolute deviation value A is determined according to the following formula_Hor：

Where N is the size of the current PU, N is 8, P (N, j) represents the pixel value of the N row and j column in the PU, and M_Hor(n) represents the mean of the pixel values of the nth row;

(2) the absolute vertical deviation A is determined as follows_Ver：

Where P (i, n) represents the pixel value of the ith row and n column in the prediction unit PU, M_ver(n) represents the average value of the pixels of the nth column of the prediction unit PU block;

(3) the absolute deviation value A of the lower left diagonal is determined according to the following formula_DDL：

Where P (i, j) represents the pixel value in the ith row and j column in the prediction unit PU, i ∈ [ o,1, …, N-1]，j∈[o,1,…,N-1]，M_DDL(n) represents the mean value of the pixel values of the nth row along the lower left diagonal direction;

(4) for absolute deviation value A of upper left diagonal_DULThe calculation is determined by rotating the prediction unit PU by 90 degrees to the left according to the formula (5);

(5) to the horizontal upper partAbsolute deviation value A_{H_U}Horizontal deviation from the lower absolute deviation value A_{H_D}Calculating, and determining the prediction unit PU according to the formula (4) after rotating the prediction unit PU by 13 degrees leftwards or rightwards according to the angle deviation value between the mode corresponding to the prediction unit PU and the mode corresponding to the horizontal direction;

(6) to the absolute deviation value A of vertical deviation left_{V_L}Absolute deviation value A of vertical deviation to right_{V_R}Calculating, and determining the current prediction unit PU according to the formula (3) after rotating the current prediction unit PU by 13 degrees leftwards or rightwards according to the angle deviation value between the mode corresponding to the current prediction unit PU and the mode corresponding to the vertical direction;

scaling the 16 × 16 prediction unit PU and the 4 × 4 prediction unit PU into a prediction unit PU with the size of 8 × 8 by adopting a nearest neighbor interpolation method, and calculating according to a calculation method of pixel absolute deviation values ADP of the prediction unit PU with the size of 8 × 8 in each direction;

the minimum pixel absolute deviation value ADP_minThe corresponding pattern is called possible pattern PM, and the possible pattern PM and the minimum pixel absolute deviation ADP_minThe corresponding relationship of (a) is shown as follows:

constructing a set of candidate modes M for coarse mode decision from the possible modes PM of a prediction unit PU_SSelecting a possible mode PM and 6 modes left and right adjacent to the possible mode PM, and forming a mode set M for coarse mode decision together with a mode 0 and a mode 1_S；

3. The 360 degree video intra prediction mode decision method of claim 1, wherein in (2), the modes for rate distortion optimization are:

set M of candidate patterns_SDetermining 8 hadamard transforms according to equation (1)Absolute sum SATD, 2 modes with the smallest values are selected from the 16 × 16 prediction units PU, and 3 modes with the smallest values are selected from the 8 × 8 and 4 × 4 prediction units PU.

4. The 360 degree video intra prediction mode decision method according to claim 1, wherein in (2), the determined optimal mode is:

and adding the MPMS into the rate-distortion optimized mode, determining the rate-distortion cost RDcost according to the formula (2), and finding out the mode with the minimum cost value as the optimal mode.

Technical Field

The invention belongs to the technical field of video coding, and particularly relates to a 360-degree video intra-frame prediction mode decision.

Background

360-degree video is widely applied to immersive cinemas, VR games, social media, television broadcasting, remote education and the like as a virtual reality technology. The resolution of 360 degree video is typically 4K, 6K or 8K, and the frame resolution of video is typically 60 frames per second, sometimes even up to 90 frames per second, so high quality 360 degree video can consume many giga-bytes of bandwidth per second. How to effectively increase the encoding speed is a key factor for promoting the development of 360-degree video technology. Organizations such as joint image experts group, motion picture experts group, and joint video research group are performing intensive research on 360-degree videos to determine a uniform coding standard.

The processing flow of the 360-degree video comprises the processes of splicing processing, projection conversion, encoding, transmission, decoding, video rendering, back projection and the like of the 360-degree video. The projection conversion mode is a key factor for determining the encoding efficiency of the 360-degree video, and a common projection mode is equal-rectangular projection which is intuitive and easy to generate but generates oversampling near a pole, so that larger shape distortion and encoding bit waste are caused. Cube projection projects a spherical surface onto a cube surface and is arranged in a flexible manner, different arrangements leading to different compression efficiencies. Compared with equal rectangular projection, cubic projection improves the efficiency of motion estimation and motion compensation, but still has the problem of oversampling, and compared with the original spherical surface, the oversampling rate is as high as 190%. In addition, there are 11 projection modes such as equiangular cube projection, octahedron projection, spherical surface division projection, rotating spherical surface projection, equatorial cylindrical projection, icosahedron projection, etc. Different projection modes sample different areas of the sphere with different sampling densities, and no matter which projection mode is used, picture distortion to a certain degree can be brought. Most of the current 360-degree video coding optimization research is performed in combination with high-efficiency video coding. In order to improve the performance of 360-degree video coding, many works have proposed many optimization methods from the aspects of projection mode, coding framework, and intra-frame coding algorithm, etc. to improve the quality and efficiency of 360-degree video coding.

The high resolution of 360-degree video enables the CU partitioning and mode decision process in intra-frame coding to have higher coding complexity and consume more coding time than the conventional video, so that the optimization of the intra-frame coding algorithm is an important research task.

Compared with the conventional video, the projected 360-degree video generates a certain degree of distortion, which causes the encoder to exhibit different coding performance in different regions of the frame. Moreover, the high resolution of 360-degree video can make the mode decision process in intra-frame coding consume a large amount of coding time.

Disclosure of Invention

The technical problem to be solved by the present invention is to overcome the above-mentioned shortcomings of the prior art, and to provide a 360 degree video intra prediction mode decision method with low coding complexity and short coding time.

The technical scheme adopted for solving the technical problems comprises the following steps:

(1) decision-making method for constructing large-size prediction unit mode

The prediction unit PU has a size of 64 × 64, 32 × 32, and a two-stage coarse mode decision method is adopted for the current prediction unit PU:

1) coarse mode decision method for constructing first stage

The 8 prediction modes 0, 1, 2, 10, 18, 26, 30, 34 are selected for the coarse mode decision in the first stage, and the absolute values of the hadamard transforms and SATD for the 8 prediction modes are determined as follows:

h is a normalized NxN-order Hadamard matrix, N is the size of the matrix, N is 4 or 8, X is a prediction residual matrix corresponding to a prediction mode and has the same size as H, p represents a row of HXH, and q represents a column of HXH.

Two minimum patterns are selected from the absolute values and SATD of the 8 prediction modes, which are respectively called as first patterns FM₁Second mode SM₁。

2) Coarse mode decision method for constructing second stage

The mode set M for the second stage coarse mode decision RMD is constructed as follows:

the coarse mode decision of the second stage is made with mode set M.

3) Determining the absolute value and SATD of 4 Hadamard transform according to formula (1) in the candidate mode set M, and selecting two modes FM with the minimum absolute value and SATD of Hadamard transform₁And SM₁Constructing a set of modes M for rate-distortion optimized RDO using the following equation_R：

4) Determining an optimal pattern

Adding the most probable mode MPM to the mode set M_RIn the method, rate-distortion cost RDcost is determined according to equation (2), and the mode with the minimum cost value is found to be the optimal mode.

RDcost＝S_mode+λ×B'_mode (2)

Wherein, Smode represents the sum of squares of residuals between an original pixel block and a reconstructed block of the prediction unit PU, λ is a coding coefficient, and B' mode represents the number of bits consumed by intra-coding of the prediction unit PU.

(2) Decision-making method for constructing small-size prediction unit mode

The size of the prediction unit PU is 16 multiplied by 16, 8 multiplied by 8 and 4 multiplied by 4, the texture direction of the prediction unit PU is preliminarily analyzed by a pixel absolute deviation method to obtain a matched angle mode, the angle mode adjacent to the angle mode, the mode 0 and the mode 1 form a mode for rough mode decision, the absolute value of Hadamard transform of each mode and SATD are compared to obtain an adaptive mode screening result, the adaptive mode screening result and the most probable mode MPM form a mode for rate distortion optimization, a rate distortion cost value RDcost is determined according to the formula (2) by a rate distortion optimization method, and the optimal mode is selected.

In step (2) of the present invention, the method for constructing a mode for coarse mode decision is as follows:

the mode set for determining the coarse mode decision is as follows:

determining pixel absolute deviation values ADP of all pixel points in the current prediction unit PU in 8 texture directions by using a pixel absolute deviation method, wherein the pixel absolute deviation values ADP of the 8 prediction directions respectively use horizontal absolute deviation values A_HorAbsolute vertical deviation value A_VerAbsolute deviation value of lower left diagonal A_DDLAbsolute deviation value of upper left diagonal A_DULAbsolute deviation value A of horizontal deviation from the upper_{H_U}Horizontal deviation from the lower absolute deviation value A_{H_D}Absolute deviation value A of vertical deviation from left_{V_L}Absolute deviation value A of vertical deviation to right_{V_R}And (4) showing.

(1) The horizontal absolute deviation value A is determined according to the following formula_Hor：

Where N is the size of the current PU, N is 8, P (N, j) represents the pixel value of the N row and j column in the PU, and M_Hor(n) represents the mean value of the pixel values of the nth row.

(2) The absolute vertical deviation A is determined as follows_Ver：

Where P (i, n) represents the pixel value of the ith row and n column in the prediction unit PU, M_ver(n) represents the average value of the pixels of the nth column of the prediction unit PU block.

(3) The absolute deviation value A of the lower left diagonal is determined according to the following formula_DDL：

Where P (i, j) represents the pixel value in the ith row and j column in the prediction unit PU, i ∈ [ o,1 ],. N-1]，j∈[o,1,···,N-1]，M_DDL(n) represents the average of the pixel values of the nth row extending in the lower left diagonal direction.

(4) For absolute deviation value A of upper left diagonal_DULThe calculation is determined by equation (5) by rotating the prediction unit PU 90 ° to the left.

(5) To the absolute deviation value A of horizontal deviation_{H_U}Horizontal deviation from the lower absolute deviation value A_{H_D}And calculating, and determining the prediction unit PU according to the formula (4) after rotating the prediction unit PU by 13 degrees leftwards or rightwards according to the angle offset value between the mode corresponding to the prediction unit PU and the mode corresponding to the horizontal direction.

(6) To the absolute deviation value A of vertical deviation left_{V_L}Absolute deviation value A of vertical deviation to right_{V_R}Calculating the angle deviation value between the mode corresponding to the calculation result and the mode corresponding to the vertical direction,the current prediction unit PU is rotated by 13 ° to the left or right and then determined according to equation (3).

The prediction units PU of 8 × 8 are scaled from the 16 × 16 prediction units PU and the 4 × 4 prediction units PU by nearest neighbor interpolation, and the pixel absolute deviation values ADP of the prediction units PU of 8 × 8 in each direction are calculated.

The minimum pixel absolute deviation value ADP_minThe corresponding pattern is called possible pattern PM, and the possible pattern PM and the minimum pixel absolute deviation ADP_minThe corresponding relationship of (a) is shown as follows:

constructing a set of candidate modes M for coarse mode decision from the possible modes PM of a prediction unit PU_SSelecting a possible mode PM and 6 modes left and right adjacent to the possible mode PM, and forming a mode set M for coarse mode decision together with a mode 0 and a mode 1_S。

In step (2) of the present invention, the modes for rate distortion optimization are:

set M of candidate patterns_SAnd (2) determining absolute values and SATD of 8 Hadamard transform according to the formula (1), selecting 2 modes with the minimum value in a prediction unit PU of 16 × 16, and selecting 3 modes with the minimum value in prediction units of 8 × 8 and 4 × 4.

In step (2) of the present invention, the determining the optimal mode is:

and adding the MPMS into the rate-distortion optimized mode, determining the rate-distortion cost RDcost according to the formula (2), and finding out the mode with the minimum cost value as the optimal mode.

The method adopts the correlation among the size of a prediction unit PU, prediction modes and texture characteristics, provides that the size of the prediction unit PU is divided into two categories and different mode decision methods are used, a probability statistical analysis method is used for the large-size prediction unit PU, 8 modes with higher mode occupation are selected for the first-stage rough mode decision, and a mode set for the second-stage rough mode decision is constructed according to two modes with the minimum absolute value of Hadamard transform screened by the first-stage rough mode decision; the method for analyzing texture direction is adopted for the small-size prediction unit PU, 35 prediction modes are divided into 8 groups according to 8 main texture directions, the mode matching of the current prediction unit PU in the 8 main texture directions is realized by using the ADP method, the mode group corresponding to the texture direction with the minimum ADP value is selected to enter a rough mode decision process, and the selection of the optimal prediction mode is completed by combining MPM and RDO methods. The proposed algorithm can reduce the number of modes traversed by the coarse mode decision and the RDO, and realize the fast mode decision of the prediction unit PU, thereby reducing the time loss.

The method analyzes the texture characteristics of the CMP-based 360-degree video, improves the mode decision part in the intra-frame coding process, reduces the coding complexity and saves the coding time.

The invention has the following beneficial effects:

the invention fully considers the texture characteristics of the 360-degree video in the CMP format, improves the intra-coding process, and can ensure the coding efficiency while reducing the coding complexity.

The sizes of the prediction units PU are divided into two sizes, and the mode pre-screening is carried out on the PUs with different sizes according to the characteristics of the PUs, so that the number of modes traversed by a rough mode decision and a Rate Distortion Optimization (RDO) process is effectively reduced, and the quick decision of the optimal prediction mode of the prediction units PU is realized.

For the coding results of 18 360-degree video test sequences proposed by the joint video development organization, the average coding time of the proposed method is reduced by 26.4%, the average BDBR increase is 0.551%, the BDWS-PSNR is reduced by 0.07db, and the WS-PSNR is reduced by 0.0153 db. The optimization effect is obviously improved compared with the prior optimization technology.

Drawings

FIG. 1 is a flowchart of example 1 of the present invention.

Detailed Description

The present invention will be described in further detail below with reference to the drawings and examples, but the present invention is not limited to the embodiments described below.

Example 1

The 360-degree video intra prediction mode decision method of the present embodiment is composed of the following steps (see fig. 1):

(1) decision-making method for constructing large-size prediction unit mode

The prediction unit PU has a size of 64 × 64, 32 × 32, and a two-stage coarse mode decision method is adopted for the current prediction unit PU:

1) coarse mode decision method for constructing first stage

The 8 prediction modes 0, 1, 2, 10, 18, 26, 30, 34 are selected for the coarse mode decision in the first stage, and the absolute values of the hadamard transforms and SATD for the 8 prediction modes are determined as follows:

where H is a normalized nxn-order hadamard matrix, N is a size of the matrix, N is 4 in this embodiment, X is a prediction residual matrix corresponding to the prediction mode, the size is N, q represents a row of HXH, and p represents a column of HXH.

The minimum two modes are selected from the absolute value and SATD of the Hadamard transform of 8 prediction modes, which are called as a first mode and a second mode FM₁、SM₁。

2) Coarse mode decision method for constructing second stage

The mode set M for the second stage coarse mode decision RMD is constructed as follows:

the coarse mode decision of the second stage is made with mode set M.

3) Determining 4 Hadamard transforms from the candidate mode set M according to equation (1)And selecting two modes FM with the smallest sum of the absolute value of the Hadamard transform and the SATD₁And SM₁Constructing a set of modes M for rate-distortion optimized RDO using the following equation_R：

4) Determining an optimal pattern

Adding the most probable mode set MPM to the mode set M_RIn the method, rate-distortion cost RDcost is determined according to equation (2), and the mode with the minimum cost value is found to be the optimal mode.

RDcost＝S_mode+λ×B'_mode (2)

Wherein S is_modeDenotes the sum of squares of the residuals between the original pixel block and the reconstructed block of the prediction unit PU, λ is the coding coefficient, B'_modeIndicating the number of bits consumed for intra coding of the prediction unit PU.

(2) Decision-making method for constructing small-size prediction unit mode

The size of the prediction unit PU is 16 multiplied by 16, 8 multiplied by 8 and 4 multiplied by 4, the texture direction of the prediction unit PU is preliminarily analyzed by a pixel absolute deviation method to obtain a matched angle mode, the angle mode adjacent to the angle mode, the mode 0 and the mode 1 form a mode for rough mode decision, the absolute value of Hadamard transform of each mode is compared to obtain an adaptive mode screening result, the mode for rate distortion optimization is formed by the angle mode and the most probable mode MPM, a rate distortion cost value RDcost is determined according to the formula (2) by a rate distortion optimization method, and the optimal mode is selected.

The method for constructing the mode for the coarse mode decision comprises the following steps:

the mode set for determining the coarse mode decision is as follows:

determining pixel absolute deviation values ADP of all pixel points in the current prediction unit PU in 8 texture directions by using a pixel absolute deviation method, wherein the pixel absolute deviation values ADP of the 8 prediction directions respectively use horizontal absolute deviation values A_HorAbsolute vertical deviation value A_VerAbsolute deviation value of lower left diagonal A_DDLAbsolute deviation value of upper left diagonal A_DULAbsolute deviation value A of horizontal deviation from the upper_{H_U}Horizontal deviation from the lower absolute deviation value A_{H_D}Absolute deviation value A of vertical deviation from left_{V_L}Absolute deviation value A of vertical deviation to right_{V_R}And (4) showing.

1) The horizontal absolute deviation value A is determined according to the following formula_Hor：

Where N is the size of the current PU, N is 8, P (N, j) represents the pixel value of the N row and j column in the PU, and M_Hor(n) represents the mean value of the pixel values of the nth row.

2) The absolute vertical deviation A is determined as follows_Ver：

Where P (i, n) represents the pixel value of the ith row and n column in the prediction unit PU, M_ver(n) represents the average value of the pixels of the nth column of the prediction unit PU block.

3) The absolute deviation value A of the lower left diagonal is determined according to the following formula_DDL：

Where P (i, j) represents the pixel value in the ith row and j column in the prediction unit PU, i ∈ [ o,1 ],. N-1]，j∈[o,1,···,N-1]In this embodiment, the value of N is 8, M_DDL(n) represents the average of the pixel values of the nth row extending in the lower left diagonal direction.

4) For absolute deviation value A of upper left diagonal_DULThe calculation is determined by equation (5) by rotating the prediction unit PU 90 ° to the left.

5) To the absolute deviation value A of horizontal deviation_{H_U}Horizontal deviation from the lower absolute deviation value A_{H_D}And calculating, and determining the prediction unit PU according to the formula (4) after rotating the prediction unit PU by 13 degrees leftwards or rightwards according to the angle offset value between the mode corresponding to the prediction unit PU and the mode corresponding to the horizontal direction.

6) To the absolute deviation value A of vertical deviation left_{V_L}Absolute deviation value A of vertical deviation to right_{V_R}And (4) calculating, and determining the current prediction unit PU according to the formula (3) after rotating the current prediction unit PU by 13 degrees leftwards or rightwards according to the angle offset value between the mode corresponding to the current prediction unit PU and the mode corresponding to the vertical direction.

The prediction units PU of 8 × 8 are scaled from the 16 × 16 prediction units PU and the 4 × 4 prediction units PU by nearest neighbor interpolation, and the calculation is performed according to the calculation method of the pixel absolute deviation ADP of the prediction units PU of 8 × 8 in each direction.

The minimum pixel absolute deviation value ADP_minThe corresponding pattern is called possible pattern PM, and the possible pattern PM and the minimum pixel absolute deviation ADP_minThe corresponding relationship of (a) is shown as follows:

constructing a set of candidate modes M for coarse mode decision from the possible modes PM of a prediction unit PU_SSelecting a possible mode PM and 6 modes left and right adjacent to the possible mode PM, and forming a mode set M for coarse mode decision together with a mode 0 and a mode 1_S：

The modes for rate distortion optimization are as follows: set M of candidate patterns_SAnd (2) determining absolute values and SATD of 8 Hadamard transform according to the formula (1), selecting 2 modes with the minimum value in a prediction unit PU of 16 × 16, and selecting 3 modes with the minimum value in prediction units of 8 × 8 and 4 × 4.

The determined optimal mode is as follows: adding a most probable mode set MPM (disclosed in the New Generation high efficiency video coding (H.265/HEVC) principle, standard and implementation) into a rate distortion optimized mode, determining a rate distortion cost RDcost according to an equation (2), and finding out a mode with the minimum cost value, namely the optimal mode.

And finishing the 360-degree video intra-frame prediction mode decision method.

Example 2

The 360-degree video intra prediction mode decision method of the embodiment comprises the following steps:

(1) decision-making method for constructing large-size prediction unit mode

The 8 prediction modes 0, 1, 2, 10, 18, 26, 30, 34 are selected for the coarse mode decision in the first stage, and the absolute values of the hadamard transforms and SATD for the 8 prediction modes are determined as follows:

where H is a normalized nxn-order hadamard matrix, N is a size of the matrix, N of this embodiment is 8, X is a prediction residual matrix corresponding to the prediction mode, the size is N, q represents a row of HXH, and p represents a column of HXH.

The minimum two modes are selected from the absolute value and SATD of the Hadamard transform of 8 prediction modes, which are called as a first mode and a second mode FM₁Second mode SM₁。

The other steps of this step are the same as in the example 1.

The other steps are the same as those in the embodiment 1, and the 360-degree video intra prediction mode decision method is completed.

In order to verify the beneficial effects of the present invention, the inventor uses the 360-degree video intra-frame prediction mode decision method of embodiment 1 of the present invention and the existing standard 360-degree video coding algorithm HM16.16-360lib4.0 (referred to as comparative test method for short) to perform comparative simulation experiments, and the experimental conditions are as follows:

the experiment is a simulation experiment on the basis of HM16.16-360Lib4.0, wherein a hardware platform is Intel (R) core (TM) i7-6820HK CPU @2.70GGHZ, a 4-core processor, a memory is 16GB, a display card is GeForce GTX 980M, and an operating system is 64-bit Win10 flagship edition. The compiled debugging software is Visual Studio 2017, the configuration parameters are full I-frame coding mode, and QP is set to {22, 27, 32, 37 }. The experiment used 16 standard 360 degree video sequences provided by jvt containing three different resolutions of a (4K), B (6K) and C (8K) as shown in the table.

TABLE 1 video sequences used in the experiments

The comparison test results of the method and the comparison test method are shown in table 2, WS-PSNR represents weighted spherical uniform peak signal-to-noise ratio for evaluating objective quality of spherical video, BDBR represents code rate variation conditions of two different algorithms under the condition that WS-PSNR is the same, BDWS-PSNR represents the difference of WS-PSNR of the method and the comparison test method under the condition that code rates are the same, Δ T represents the average time percentage saved by the method relative to the comparison test method, and the calculation formula is as follows:

wherein, Time_HM(i) Time for the comparison of the coding Time of the test method under the i-th quantization parameter_prop(i) The coding time of the inventive method under the i-th quantization parameter. Delta WS-PSNR is the ratio of the method of the present invention to the comparative test method in coding bitsUnder the condition of the same rate, the amount of change of the video quality is calculated according to the following formula:

wherein WS-PSNR_HM(i) Encoding video quality, WS-PSNR, of reconstructed video under ith quantization parameter for comparative experimental methods_prop(i) The video quality of the reconstructed video is encoded for the method of the invention under the ith quantization parameter.

TABLE 2 comparison of the Performance of the present algorithm with HM16.16-360Lib4.0

The experimental results in table 2 show that, compared with the comparative test method, the average encoding time of the invention is reduced by 26.4%, the average BDBR is increased by 0.551%, the BDW-SPSNR is reduced by 0.07db, the Delta WS-PSNR is reduced by 0.0153db, and the encoding performance loss is negligible. For different types of video sequences, the method can effectively reduce the coding time, improve the coding efficiency and ensure the video quality. The method is suitable for different test sequences, has obvious optimization effect and good theoretical significance and application value.