CN108366256A - A kind of HEVC intra prediction modes quickly select system and method - Google Patents

Abstract

The present invention proposes a HEVC intra-frame prediction mode fast selection system and method, which are used to reduce the complexity of HEVC intra-frame prediction coding mode selection; the system includes five modules, and the MPM mode determination module determines three MPMs for the current prediction block Prediction mode; the mode area division module divides all prediction directions into three areas; the mode direction matching module obtains a high-probability prediction mode candidate list; the rate-distortion cost calculation module calculates the RDcost of each mode in the high-probability prediction mode candidate list to obtain RDcost Set; the optimal mode selection module sorts the RDcost set in descending order, and selects the mode corresponding to the minimum RDcost value as the optimal prediction mode. The present invention has the characteristics of high coding efficiency and high prediction accuracy, and can be used for intra-frame prediction in the HEVC video standard coding.

Description

A HEVC intra prediction mode fast selection system and method

technical field

The invention belongs to the technical field of digital signal processing, and in particular relates to a system and method for quickly selecting an HEVC intra-frame prediction mode, which can be used for intra-frame prediction coding in the HEVC video standard.

Background technique

HEVC is a new video compression standard used to replace the H.264/AVC coding standard. On January 26, 2013, HEVC officially became an international standard. As a new-generation video coding standard, HEVC (H.265) still belongs to the hybrid coding framework of prediction plus transformation. However, compared to H.264, H.265 has revolutionary changes in many aspects.

In H.265, the size of the macroblock is extended from 16×16 in H.264 to 64×64, and predictive coding, transform coding and entropy coding are performed on the image to be coded sequentially. Among them, predictive coding needs to select the best coding unit CU, and select the prediction mode of each prediction unit PU in the CU. There are 35 prediction modes of the PU, including Planar mode (0 mode), DC mode (1 mode) and 33 angle prediction mode.

In the HEVC predictive encoding process, the coding unit CU is selected first, and then the optimal prediction mode is selected for the prediction unit PU, that is, the current prediction block. In the traditional technology of selecting the optimal prediction mode for the prediction unit PU, the MPM mode determination module determines three MPM prediction modes with the highest probability of becoming the optimal prediction mode of the current prediction block for the current prediction block, and the prediction mode traversal module traverses 35 types Predict the mode and calculate the absolute error and SATD of each mode to obtain a candidate list of high-probability forecast modes. The rate-distortion cost calculation module integrates the candidate list of high-probability forecast modes and three MPM forecast modes to obtain the optimal forecast mode list, and Calculate the rate-distortion cost RDcost of each mode in the optimal prediction mode list, obtain the cost RDcost set, and then select the prediction mode corresponding to the minimum value from the cost RDcost set, which is the prediction mode of the current prediction block.

Each coding unit CU in HEVC can be divided into several types of prediction units PU. When each prediction unit PU is used as the current prediction block, if the optimal prediction mode is selected, 35 prediction modes must be traversed one by one, which will lead to optimal prediction. The complexity of mode selection is very high, which greatly reduces the coding efficiency of HEVC.

Contents of the invention

The purpose of the present invention is to overcome the above-mentioned defects in the prior art, and provide a HEVC intra-frame prediction mode fast selection system and method for reducing the complexity of HEVC intra-frame prediction coding mode selection.

In order to achieve the above object, the technical scheme that the present invention takes is:

A fast selection system for HEVC intra-frame prediction mode, including sequentially connected MPM mode determination module, rate-distortion cost calculation module and optimal mode selection module, wherein:

The MPM mode determination module is used to determine three MPM prediction modes with the highest probability of becoming the optimal prediction mode of the current prediction block for the current prediction block;

The rate-distortion cost calculation module is used to obtain an optimal mode candidate set, and calculate the rate-distortion cost RDcost of each mode in the optimal mode candidate set to obtain a cost RDcost set;

The optimal prediction mode selection module is used to sort the cost RDcost set in descending order, and select the smallest cost RDcost to obtain the optimal prediction mode of the current prediction block;

Also includes a sequentially connected pattern region division module and pattern direction matching module, where:

The mode area division module is used to divide the 33 angle prediction modes in HEVC according to the direction;

The mode direction matching module is used to use the prediction modes of the left and upper blocks of the current prediction block obtained by the MPM mode determination module to perform direction matching on the three regions obtained by the mode region division module of the current prediction block, so as to obtain high A list of probabilistic prediction mode candidates.

A method for fast selection of HEVC intra-frame prediction mode, comprising the steps of:

(1) The MPM mode determination module determines three MPM prediction modes:

(1a) The MPM mode determination module records the first prediction mode of the MPM prediction mode as ModeA, the second prediction mode as ModeB, and the third prediction mode as ModeC, and the prediction mode value of the left prediction block of the current prediction block Assign the value to ModeA, and assign the prediction mode value of the upper side prediction block to ModeB;

(1b) The MPM mode determination module judges whether ModeA and ModeB are equal, if so, when ModeA is 0 or 1, assign any two values in 0, 1 and 26 to ModeB and ModeC respectively, when ModeA is any of 2 to 34 When it is a value, assign the two angle prediction modes adjacent to ModeA to ModeB and ModeC respectively, otherwise, set ModeC to 0, 1 or 26, and it is not repeated with ModeA and ModeB;

(2) The mode area division module divides the 33 kinds of angle prediction modes in HEVC into mode areas, and obtains three prediction areas:

The model area division module divides 33 kinds of angle prediction modes in HEVC into the first prediction area Area1, the second prediction area Area2 and the third prediction area Area3, and the total number of prediction modes contained in each area is an odd number;

(3) The pattern direction matching module obtains the high probability prediction pattern candidate list:

(3a) mode direction matching module judges whether ModeA and ModeB belong to Area1, one of Area2 and Area3 three areas, if so, the prediction mode between ModeA and ModeB is used as high probability prediction mode candidate list, otherwise, execution step (3b );

(3b) The pattern direction matching module sets the matching interval, and records the left end point of the matching interval as the left end point ref1, the right end point as the right end point ref2, and the middle position as the middle point Area(n)_ref, where n represents prediction mode value, and 2≤n≤34;

(3c) The mode direction matching module assigns ModeA to ref1, ModeB to ref2, and (ref1+ref2)/2 to Area(n)_ref;

(3d) pattern direction matching module calculates the pattern matching value MSE _ref of Area(n)_ref, the pattern matching value MSE ₁ of ref1 and the pattern matching value MSE ₂ of ref2;

(3e) The mode direction matching module judges whether MSE ₁ ≤ MSE _ref ≤ MSE ₂ and |ref1-ref2|≤5 are established at the same time, if so, perform step (3g), otherwise, perform step (3f);

(3f) The mode direction matching module assigns Area(n)_ref to ref1, (Area(n)_ref+ref2)/2 assigns to Area(n)_ref, ModeB assigns to ref2, and executes step (3d);

(3g) the mode direction matching module uses the prediction mode between ref1 and ref2 as a high probability prediction mode candidate list;

(4) The rate-distortion cost calculation module obtains the optimal mode candidate list, and calculates the rate-distortion cost RDcost of each mode in the optimal mode candidate list, and obtains the cost RDcost set:

(4a) The rate-distortion cost calculation module integrates the three MPM prediction modes determined in step (1) with the high-probability prediction mode candidate list obtained in step (3), to obtain the optimal prediction mode candidate list;

(4b) The rate-distortion cost calculation module calculates the rate-distortion cost RDcost of each mode in the optimal prediction mode candidate list, and obtains a set of cost RDcost;

(5) The optimal prediction mode selection module obtains the optimal prediction mode of the current prediction block:

The optimal prediction mode selection module sorts the RDcost sets calculated by the rate-distortion cost calculation module in descending order, and selects the prediction mode corresponding to the smallest cost RDcost value as the optimal prediction mode of the current prediction block.

Compared with traditional technology, the present invention has following advantages:

1. When the present invention obtains the high-probability prediction mode candidate list, it first uses the mode area division module to divide the 33 prediction modes in HEVC into three types of prediction areas, and then uses the mode direction matching module and the MPM mode determination module to determine For the prediction mode of the left and upper blocks of the current prediction block, perform direction matching on the three areas divided by the mode area division module of the current prediction block, quickly locate and narrow the matching interval, and then obtain a high-probability prediction mode candidate list, which is different from the existing technology Compared with obtaining the high-probability prediction mode candidate list by traversing the 35 prediction modes one by one through the prediction mode traversal module, the number of calculations is reduced, the coding complexity is reduced, and the coding efficiency is effectively improved.

2. When the present invention divides the 33 kinds of prediction modes in HEVC, the mode area division module is used, and according to the directional correlation between each prediction mode, the 33 kinds of angle prediction modes are divided into three kinds of prediction areas, namely The positive offset area in the vertical direction, the positive offset area in the horizontal direction, and the negative offset area in the vertical and horizontal directions, and the total number of prediction modes contained in each prediction area is odd, which provides convenience for the subsequent direction matching process and further improves the coding efficiency

3. When calculating the mode matching value MSE, the present invention uses a mode direction matching module to calculate the two-dimensional mean square error between the current prediction block and the reference block pixel value, compared to the method of calculating absolute error and SAD in the traditional method , which reduces the prediction error and has the characteristics of higher prediction accuracy.

Description of drawings

Fig. 1 is the overall structure schematic diagram of the selection system of the present invention;

Fig. 2 is a flow chart of the realization of the selection method of the present invention.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments.

Referring to Fig. 1, a method for fast selection of an HEVC intra-frame prediction mode includes the following steps:

A fast selection system for HEVC intra-frame prediction mode, including sequentially connected MPM mode determination module, rate-distortion cost calculation module and optimal mode selection module, wherein:

The MPM mode determination module is used to determine three MPM prediction modes with the highest probability of becoming the optimal prediction mode of the current prediction block for the current prediction block;

The rate-distortion cost calculation module is used to obtain an optimal mode candidate set, and calculate the rate-distortion cost RDcost of each mode in the optimal mode candidate set to obtain a cost RDcost set;

The optimal prediction mode selection module is used to sort the cost RDcost set in descending order, and select the smallest cost RDcost to obtain the optimal prediction mode of the current prediction block;

Also includes a sequentially connected pattern region division module and pattern direction matching module, where:

The mode area division module is used to divide the 33 angle prediction modes in HEVC according to the direction;

The mode direction matching module is used to use the prediction modes of the left and upper blocks of the current prediction block obtained by the MPM mode determination module to perform direction matching on the three regions obtained by the mode region division module of the current prediction block, so as to obtain High probability prediction mode candidate list.

Referring to Fig. 2, a method for fast selection of an HEVC intra prediction mode includes the following steps:

Step 1) MPM mode determination module determines three MPM prediction modes:

Step 1a) The MPM mode determination module records the first prediction mode of the MPM prediction mode as ModeA, the second prediction mode as ModeB, and the third prediction mode as ModeC, and the prediction mode value of the left prediction block of the current prediction block Assign the value to ModeA, and assign the prediction mode value of the upper side prediction block to ModeB;

Step 1b) MPM mode determines whether the module judges whether ModeA and ModeB are equal, if so, when ModeA is 0 or 1, assign any two values in 0, 1 and 26 to ModeB and ModeC respectively, when ModeA is any of 2 to 34 When it is a value, assign the two angle prediction modes adjacent to ModeA to ModeB and ModeC respectively, otherwise, set ModeC to 0, 1 or 26, and it is not repeated with ModeA and ModeB;

Step 2) The mode area division module divides the mode area into 33 kinds of angle prediction modes in HEVC, and obtains three prediction areas:

The mode area division module divides the 33 angle prediction modes into vertical positive offset area Area1 and horizontal direction positive offset area Area2 according to the positive and negative offset values of the 33 angle prediction modes in HEVC in the horizontal and vertical directions There are three areas together with the negative offset area Area3 in the vertical and horizontal directions. Area1 indicates the area with positive offset in the vertical direction, and is predicted to the upper right direction of the current block through Area1. Area2 indicates the area with positive offset in the horizontal direction. Use Area2 to predict the lower-left direction of the current block, and Area3 indicates an area with negative offsets in the vertical and horizontal directions, and use Area3 to predict the lower-right direction of the current block, and the total number of prediction modes contained in each area is an odd number, among them, Area1 Including 2 to 10, Area2 includes 11 to 27, Area3 includes 28 to 34;

Step 3) The pattern direction matching module obtains the high probability prediction pattern candidate list:

Step 3a) Mode direction matching module judges whether ModeA and ModeB belong to Area1, one of Area2 and Area3 three areas, if so, the prediction mode between ModeA and ModeB is used as the high probability prediction mode candidate list, otherwise, execute step 3b) ;

Step 3b) The pattern direction matching module sets the matching interval, and records the left end point of the matching interval as the left end point ref1, the right end point as the right end point ref2, and the middle position as the middle point Area(n)_ref, where n represents prediction mode value, and 2≤n≤34;

Step 3c) The mode direction matching module assigns ModeA to ref1, ModeB to ref2, and (ref1+ref2)/2 to Area(n)_ref;

Step 3d) The pattern direction matching module calculates the pattern matching value MSE _ref of Area(n)_ref, the pattern matching value MSE ₁ of ref1 and the pattern matching value MSE ₂ of ref2, and the pattern matching value is obtained by a two-dimensional mean square error method, and the two-dimensional mean The formula for calculating the variance is:

Among them, M and N represent the width and height of the current prediction block respectively, f _i and f _i-1 represent the pixel values of the current block and the reference block respectively, x and y represent the horizontal components of the motion vectors corresponding to the pixel positions of the current block and the reference block and the vertical component;

Step 3e) The pattern direction matching module judges whether MSE ₁ ≤ MSE _ref ≤ MSE ₂ and |ref1-ref2|≤5 are all established, if so, perform step (3g), otherwise, perform step (3f);

Step 3f) Mode direction matching module assigns Area(n)_ref to ref1, (Area(n)_ref+ref2)/2 assigns to Area(n)_ref, ModeB assigns to ref2, and executes step 3d);

Step 3g) The mode direction matching module uses the prediction mode between ref1 and ref2 as a high-probability prediction mode candidate list;

Step 4) The rate-distortion cost calculation module obtains the optimal mode candidate list and calculates the rate-distortion cost RDcost of each mode in the optimal mode candidate list, and obtains the cost RDcost set:

Step 4a) The rate-distortion cost calculation module integrates the three MPM prediction modes obtained in step 1) with the high-probability prediction mode candidate list obtained in step 3), to obtain the optimal prediction mode candidate list;

Step 4b) The rate-distortion cost calculation module calculates the rate-distortion cost RDcost of each mode in the optimal prediction mode candidate list, and obtains the cost RDcost set, wherein the rate-distortion cost RDcost is defined according to the rate-distortion cost value defined in HEVC. Each mode is calculated, and the calculation formula is:

RDcost＝Distortion+λ×uiBits

Among them, Distortion is the sum of squared errors between the reconstructed image and the original image after prediction using the prediction mode, λ is the Lagrangian factor, and uiBits is generated by sequentially predicting, transforming, quantizing, and entropy encoding the image using the prediction mode number of encoded bits;

Step 5) The optimal prediction mode selection module obtains the optimal prediction mode of the current prediction block:

The optimal prediction mode selection module sorts the RDcost sets obtained by the rate-distortion cost calculation module in descending order, and selects the prediction mode corresponding to the smallest cost RDcost value as the optimal prediction mode of the current prediction block.

The above description is only a specific example of the present invention. Obviously, for those skilled in the art, after understanding the content and principle of the present invention, it is possible to carry out the form and details without departing from the principle and structure of the present invention. Various amendments and changes, but these amendments and changes based on the idea of the present invention are still within the protection scope of the claims of the present invention.

Claims (5)

1. a kind of HEVC intra prediction modes quickly select system, including sequentially connected MPM mode decision modules, rate to be distorted generation Valence computing module and optimization model choose module, wherein：

The MPM mode decision modules, for determining that three can become current prediction block optimum prediction mould for current prediction block The highest MPM prediction modes of formula probability；

The rate distortion costs computing module for obtaining optimization model candidate collection, and calculates the optimization model candidate collection In each pattern rate distortion costs RDcost, to obtain cost RDcost set；

The optimal prediction modes choose module, for carrying out descending sort to cost RDcost set, and choose minimum generation Valence RDcost, to obtain the optimal prediction modes of current prediction block；

It is characterized in that, further include sequentially connected mode region division module and pattern direction matching module, wherein：

The mode region division module, for carrying out region division to 33 kinds of angle prediction modes in HEVC by direction；

Pattern direction matching module, the current prediction block left side for being determined using MPM mode decision modules and upper block Prediction mode, to current prediction block three regions that mode region division module divides into line direction match, to obtain height Probability Forecast Model candidate list.

2. a kind of method that HEVC intra prediction modes quickly select, which is characterized in that include the following steps：

(1) MPM mode decision modules determine three MPM prediction modes：

First prediction mode of MPM prediction modes is denoted as ModeA by (1a) MPM mode decision modules, and the second prediction mode is denoted as ModeB, third prediction mode is denoted as ModeC, and the prediction mode value of the left side prediction block of current prediction block is assigned to The prediction mode value of ModeA, upside prediction block are assigned to ModeB；

(1b) MPM mode decision modules judge whether ModeA and ModeB is equal, if so, when ModeA is 0 or 1, by 0,1 With 26 in any two value be assigned to ModeB and ModeC respectively, will be with ModeA when ModeA is any one value in 2 to 34 Two adjacent angle prediction modes are assigned to ModeB and ModeC respectively, otherwise, ModeC are set as 0,1 or 26, and with ModeA and ModeB are not repeated；

(2) mode region division module carries out mode region division to 33 kinds of angle prediction modes in HEVC：

33 kinds of angle prediction modes in HEVC are divided into the first estimation range Area1, second in advance by mode region division module Region Area2 and third estimation range Area3 is surveyed, and the prediction mode sum that each region includes is odd number；

(3) pattern direction matching module obtains high-confidence forecast mode candidate list：

(3a) pattern direction matching module judges whether ModeA and ModeB belongs in tri- regions Area1, Area2 and Area3 One, if so, using the prediction mode between ModeA and ModeB as high-confidence forecast mode candidate list, otherwise, execute Step (3b)；

(3b) pattern direction matching module setting matching section, and the left side endpoint for matching section is denoted as left end point ref1, it is right Side point is denoted as right endpoint ref2, and centre position is denoted as intermediate point Area (n) _ ref, and wherein n represents prediction mode value, and 2≤n ≤34；

ModeA is assigned to ref1 by (3c) pattern direction matching module, and ModeB is assigned to ref2, and (ref1+ref2)/2 is assigned to Area(n)_ref；

(3d) pattern direction matching module calculates the pattern match value MSE of Area (n) _ ref_ref, ref1 pattern match value MSE₁ With the pattern match value MSE of ref2₂；

(3e) pattern direction matching module judges MSE₁≤MSE_ref≤MSE₂With | ref1-ref2 | whether≤5 set up simultaneously, if It is to execute step (3g), otherwise, executes step (3f)；

Area (n) _ ref is assigned to ref1 by (3f) pattern direction matching module, and (Area (n) _ ref+ref2)/2 is assigned to Area (n) _ ref, ModeB are assigned to ref2, and execute step (3d)；

(3g) pattern direction matching module is using the prediction mode between ref1 and ref2 as high-confidence forecast mode candidate list；

(4) rate distortion costs computing module obtains optimization model candidate list, and calculates each mould in optimization model candidate list The rate distortion costs RDcost of formula obtains cost RDcost set：

The high probability that three MPM prediction modes that (4a) rate distortion costs computing module determines step (1) are obtained with step (3) Prediction mode candidate list is integrated, and optimal prediction modes candidate list is obtained；

(4b) rate distortion costs computing module calculates the rate distortion costs of each pattern in optimal prediction modes candidate list RDcost obtains cost RDcost set；

(5) optimal prediction modes choose the optimal prediction modes that module obtains current prediction block：

Optimal prediction modes choose the RDcost set that rate distortion costs computing module is calculated module and carry out descending sort, And select prediction mode corresponding to minimum cost RDcost values, the optimal prediction modes as current prediction block.

3. a kind of method that HEVC intra prediction modes quickly select according to claim 2, it is characterised in that：Step (2) the mode region division module described in carries out mode region division to 33 kinds of angle prediction modes in HEVC, realizes Cheng Wei：

Mode region division module according to the deviant of 33 kinds of angle prediction modes in the horizontal and vertical directions in HEVC just Forsake one's love condition, by 33 kinds of angle prediction modes be divided into vertical direction positive offset area Area1, horizontal direction positive offset area Area2 and Totally three regions vertical and horizontal direction negative offset area Area3.

4. a kind of method that HEVC intra prediction modes quickly select according to claim 2, it is characterised in that：Step The pattern match value MSE of calculating Area (n) _ ref described in (3d)_ref, ref1 pattern match value MSE₁With the pattern of ref2 Matching value MSE₂, using two-dimentional mean square deviation method, the formula expression of two-dimentional mean square deviation is：

Wherein, M and N respectively represents the width and height of current prediction block, f_iAnd f_i-1The pixel value of current block and reference block is indicated respectively, X and y indicates the horizontal component and vertical component of current block and reference block respective pixel position motion vector.

5. the method that a kind of intra prediction mode based on HEVC standard according to claim 2 quickly selects, feature It is：The rate distortion costs RDcost for calculating each pattern in optimal prediction modes candidate list described in step (4b), It is to be calculated each pattern according to rate distortion costs value definition specified in HEVC, calculation formula is：

RDcost=Distortion+ λ × uiBits

Wherein, Distortion is the reconstruction image obtained after being predicted using prediction mode and the square-error of original image Lagrange factor with, λ, uiBits be image is predicted successively using prediction mode, is converted, is quantified and entropy coding after The number of coded bits of generation.