WO2017010642A1

WO2017010642A1 - Method and device for pruning prediction unit for hevc inter prediction

Info

Publication number: WO2017010642A1
Application number: PCT/KR2016/001213
Authority: WO
Inventors: 홍유표; 이은총; 강동길; 강영석
Original assignee: 동국대학교 산학협력단
Priority date: 2015-07-10
Filing date: 2016-02-04
Publication date: 2017-01-19
Also published as: KR101695769B1

Abstract

Provided is a method for pruning a PU (Prediction Unit) for HEVC inter prediction, the method being one for realizing a computer related to a PU pruning technique for HEVC (High Efficiency Video Coding), and comprising the steps of: (a) calculating a skip cost for each of a 64x64 block, a 32x32 block, and a 16x16 block that correspond to a basic unit block of a CTU (Coding Tree Block); (b) selecting 2 unit blocks to carry out PU prediction from among the three unit blocks in the order of lowest to highest skip costs, according to a comparison of each of the calculated skip costs; and (c) carrying out a PU prediction for the 2 selected unit blocks, and allocating a candidate PU partition block for the unit block among the 2 selected unit blocks which has the higher skip cost in order to inter predict the unit block having the lower skip cost, and carrying out PU prediction.

Description

Prediction unit pruning method and apparatus for HEVC inter prediction

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to high-efficiency video coding (HEVC) technology, and more particularly, to a Priction Unit Pruning Algorithm for HEVC inter prediction and an apparatus for executing the same.

As time required for high-definition video, HEVC (High-Efficiency Video Coding), an image compression standard that is advantageous for high-definition video compression, has been announced, compared to H.264, a video compression standard. There are three major differences between H.264 and HEVC.

First, the macro block (MB: Macro Block) in the existing H.264, which is a basic unit of compression, has been replaced with a coding tree unit (CTU). At this time, the size of the MB has a size of 16x16, while the size of the CTU may be selected from 16x16, 32x32, 64x64. One CTU may be further divided into sub-coding units (CUs), which are further divided into prediction units (PUs) and transform units (TUs). Here, each CU may have a size of 64x64 to 8x8, a PU may have a size of 64x64 to 4x4, and a TU may have a size of 32x32 to 4x4. Therefore, HEVC supports various sized block and frequency conversions compared to H.264. Second, while H.264 has nine prediction directions for intra-prediction, HEVC uses 35 prediction directions. Third, even in the case of inter-prediction, the maximum block size of H.264 inter prediction is limited to 16x16, while HEVC can handle various block sizes from 4x4 to 64x64. So that it was extended.

For the same reason, HEVC has a 2x improvement in image compression compared to H.264. On the other hand, compared to H.264, HEVC has a large amount of partition size to perform compression, which greatly increases the computational amount. This makes them vulnerable to real-time compression of high quality images. Therefore, an algorithm is needed to reduce excessive computation.

The present invention provides a PU pruning method and apparatus for reducing the amount of computation of HEVC inter prediction by performing prediction on the prediction unit (PU) when performing compression in performing HEVC inter prediction. To provide.

According to an aspect of the present invention, a computer-implemented method for a pruning technique of a prediction unit (PU) for high efficiency video coding (HEVC) inter prediction, comprising: (a) a coding tree block (CTU); Calculating a skip cost for each of a 64x64 block, a 32x32 block, and a 16x16 block corresponding to a basic unit block of a Coding Tree Block; (b) selecting two unit blocks to perform PU prediction among the three unit blocks in order of having a relatively low skip cost through comparing the calculated skip costs; And (c) performing PU prediction on the selected two unit blocks, wherein more candidate PU partition blocks are used for inter prediction of a unit block having a lower skip cost among the selected two unit blocks. There is provided a PU pruning method for HEVC inter prediction, which comprises performing a PU prediction by allocating ().

In one embodiment, for each of the two selected unit blocks, calculating each rate-distortion cost (RD cost) for all candidate PU partition blocks available for inter prediction. It may include.

In one embodiment, the step (c),

For inter prediction of a unit block having a lower skip cost among the two selected unit blocks, PU prediction is performed using a first number of PU partition blocks that are predetermined in the order of low rate-distortion cost for the PU partition block. Performing; And

A second number pre-specified in order of low rate-distortion cost for a PU partition block for inter prediction of a unit block having a higher skip cost among the two selected unit blocks, wherein the second number is the first number; Performing PU prediction using a PU partition block having a number smaller than 1).

In one embodiment, the rate-distortion cost for the PU partition block may be calculated by Equation 1 below.

[Equation 1]

puPairCost = ∑│puMvMedian-skipMv│ × skipRDCost

Here, puPairCost represents a rate-distortion cost for a specific PU partition block, skipMv represents a skip motion vector of each 8x8 block in the corresponding PU partition block, and puMvMedian is an intermediate of skipped motion vectors of all 8x8 blocks in the corresponding PU partition block. A median value, and skipRDCost represents a rate-distortion cost of a skip motion vector of each 8x8 block.

According to another aspect of the present invention, an apparatus to which a pruning method of a prediction unit (PU) for high efficiency video coding (HEVC) inter prediction is applied,

A skip cost calculator configured to calculate a skip cost for each of 64x64 blocks, 32x32 blocks, and 16x16 blocks corresponding to basic unit blocks of a coding tree block (CTU); A PU prediction determination unit selecting two unit blocks for performing PU prediction among the three unit blocks in an order having a relatively low skip cost by comparing the calculated skip costs; And perform PU prediction on the selected two unit blocks, and allocate more candidate PU partition blocks for inter prediction of the unit block having the lower skip cost among the selected two unit blocks. There is provided a PU prediction determination apparatus for HEVC inter prediction including a PU prediction execution unit that performs PU prediction.

In an embodiment, for each of the two selected unit blocks, an RDcost calculator for calculating a rate-distortion cost (RD cost) for all candidate PU partition blocks available for inter prediction. It may further include.

In one embodiment, the PU prediction performing unit,

For inter prediction of a unit block having a lower skip cost among the two selected unit blocks, PU prediction is performed using a first number of PU partition blocks that are predetermined in the order of low rate-distortion cost for the PU partition block. Doing,

A second number pre-specified in order of low rate-distortion cost for a PU partition block for inter prediction of a unit block having a higher skip cost among the two selected unit blocks, wherein the second number is the first number; PU prediction may be performed using a PU partition block having a smaller number than 1).

In one embodiment, the RDcost calculator may calculate the rate-distortion cost for the PU partition block by Equation 2 below.

[Equation 2]

puPairCost = ∑│puMvMedian-skipMv│ × skipRDCost

According to an embodiment of the present invention, in performing the HEVC inter prediction, when the compression is performed, the prediction unit (PU) is not performed when the compression is performed, thereby reducing the amount of computation and the computation time of the HEVC inter prediction. As a result, it is possible to reduce the size of hardware that performs compression.

1 is a flowchart of encoding of HM reference software.

2 is a schematic block diagram of an apparatus for determining PU prediction for HEVC inter prediction according to an embodiment of the present invention.

3 is a schematic flowchart of a PU pruning method for HEVC inter prediction according to an embodiment of the present invention;

4 is an exemplary diagram for explaining an RDcost prediction scheme of a PU based on a skip cost.

5 is a diagram for describing candidate PU partition blocks.

6 is a diagram illustrating a skip mode motion vector (MV) of each 8x8 blocks in a PU partition block.

7 is a table for explaining an example of the performance of the PU pruning method according to an embodiment of the present invention.

As the invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all transformations, equivalents, and substitutes included in the spirit and scope of the present invention.

In describing the present invention, when it is determined that the detailed description of the related known technology may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted. In addition, numerals (eg, first, second, etc.) used in the description process of the present specification are merely identification symbols for distinguishing one component from another component.

In addition, throughout the specification, when one component is referred to as "connected" or "connected" with another component, the one component may be directly connected or directly connected to the other component, but in particular It is to be understood that, unless there is an opposite substrate, it may be connected or connected via another component in the middle.

In addition, throughout the specification, when a part is said to "include" a certain component, it means that it may further include other components, without excluding the other components unless otherwise stated. In addition, the terms "unit", "module", and the like described in the specification mean a unit that processes at least one function or operation, which means that it may be implemented by one or more pieces of hardware or software or a combination of hardware and software. .

1 is a flowchart of encoding of HM reference software. Here, the 2N × 2N coding unit (CU) represents a CU of the largest block size (ie, 64 × 64 block size) of HEVC. Referring to FIG. 1, first, skip mode prediction is performed on a 2N × 2N CU, and conventional inter-mode prediction is performed on the same CU. And partitions of various sizes (ie, symmetric PU partition blocks (see partition blocks of square and rectangular shapes in FIG. 5) and asymmetric PU partition blocks (see partition blocks of asymmetric shapes in FIG. 5)). PU prediction is performed. As such, when the inter prediction on the CU is completed, intra-prediction is performed, and accordingly, a best mode regarding a coding tree block (CTB) is determined. This procedure is repeated for smaller CTBs.

2 is a schematic block diagram of an apparatus for determining PU prediction for HEVC inter prediction according to an embodiment of the present invention, and FIG. 3 is a schematic flowchart of a PU pruning method for HEVC inter prediction according to an embodiment of the present invention. to be.

4 is an exemplary diagram for explaining an RDcost prediction method of a PU based on a skip cost, FIG. 5 is a diagram for describing candidate PU partition blocks, and FIG. 6 is a skip mode MV of each 8x8 blocks in a PU partition block. motion vector) is an exemplary view, and FIG. 7 is a table for explaining a performance example of a PU pruning method according to an embodiment of the present invention.

Referring to FIG. 2, a PU prediction determining apparatus 200 to which a PU pruning method for HEVC inter prediction according to an embodiment of the present invention is applied may include a skip cost calculator 210, a PU prediction determiner 220, It may include an RDcost calculator 230 and a PU prediction performer 240.

With this configuration, a PU pruning method for HEVC inter prediction may be performed in the same order as in FIG. 3. In one implementation, the PU pruning method for HEVC inter prediction according to an embodiment of the present invention may be implemented by computer readable program code. In this case, the computer program related to the PU pruning method for HEVC inter prediction according to an embodiment of the present invention may be installed in a user's computer or downloaded on-line, and may be manufactured in a form that is independent or merged with another program (that is, Software products). In another implementation, the PU pruning method for HEVC inter prediction according to the embodiment of the present invention may be implemented in the form of the computer program and stored in a computer-readable recording medium. Here, the computer-readable recording medium includes all kinds of recording media storing data that can be decrypted by a computer system. For example, there may be a read only memory (ROM), a random access memory (RAM), a magnetic tape, a magnetic disk, a flash memory, an optical data storage device, and the like. The computer readable recording medium can also be distributed over computer systems connected over a computer network, stored and executed as readable code in a distributed fashion.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the block diagram of FIG. 2 and FIGS. 4 to 7 with reference to the flowchart of FIG. 3.

Referring to step S310 of FIG. 3, first, the skip cost calculator 210 calculates a skip cost for each basic unit block of the CTU. From HEVC to CTU, 64x64 blocks, 32x32 blocks, and 16x16 blocks can be selected as basic unit blocks. In this step, the skip cost for each basic unit block (ie, 64x64 blocks, 32x32 blocks, and 16x16 blocks) is calculated. .

In this case, a method of calculating the skip cost may be as follows. After calculating the Rdcost for the MV of the upper block and the left block, respectively, let the skip cost be the smaller value among the calculated Rdcost values. At this time, each MV calculates the Rdcost value in two ways. The first method is to calculate the RDcost value of MV, which is a general method. The second method is to calculate the Rdcost value after setting the residual value to 0. Since this calculation method does not apply the search range, the calculation time is short.

In step S320 of FIG. 3, a comparison of the skip cost calculated through the previous step is performed. FIG. 4 illustrates a patterned skip cost for each basic unit block calculated through the above steps. Referring to FIG. 4A, a case in which a skip cost is lowered as the block size increases, such as 64x64 blocks-> 32x32 blocks-> 16x16 blocks, is illustrated. Conversely, referring to b of FIG. 4, a case in which a skip cost is lowered as a block size decreases is illustrated. In addition, other various cases may occur, such as the cases of c to f of FIG. 4.

In this step, two basic unit blocks are selected in order of decreasing skip cost among the three (that is, 64x64 blocks, 32x32 blocks, and 16x16 blocks) by comparing the skip costs calculated for each basic unit block. This may be performed by the PU prediction determining unit 220 of the PU prediction determining apparatus 200. Thus, in the case of a of FIG. 4, as a basic unit block for performing PU prediction, a 64x64 CTB block having the lowest skip cost and a 32x32 CTB block having the next lowest skip cost will be selected, and the highest skip cost will be selected. The 16x16 CTB block with will be excluded from the PU prediction.

PU prediction is performed on two basic unit blocks selected through this step, and a specific rule according to an embodiment of the present invention is applied to the PU prediction process at this time. This will be described in steps S330 and S340 of FIG. 3.

Referring to step S330 of FIG. 3, all candidate PU partition blocks (hereinafter, referred to as candidate PU partition blocks) available for inter prediction of the corresponding basic unit blocks based on the two basic unit blocks selected through the previous steps. Compute the rate-distortion cost (RDcost) of each of The candidate PU partition block is illustrated through FIG. 5, and there may be a total of eight PU partition blocks.

The RDcost calculation for each of the candidate PU partition blocks described above may be performed by the RDcost calculator 230 of the PU prediction determining apparatus 200. In this case, the RDcost calculation for each candidate PU partition block may be calculated by the following equation.

puPairCost = ∑│puMvMedian-skipMv│ × skipRDCost

That is, in the embodiment of the present invention, in calculating the RDcost for each candidate PU partition block, the 8x8 block, which is the smallest CU leaf node block unit, uses a motion vector value. For example, a skip motion vector value, which is a motion vector in a skip mode, is used. Such a skip motion vector is illustrated in FIG. 6. For example, FIG. 6A illustrates a case in which a PU partition block is configured by pairing 8x8 blocks by 8 and configuring a pair of 32x16 blocks. According to the example of FIG. 6, when the PU partition block such as b of FIG. 6 is configured such that the direction of the motion vector in the skip mode is mostly coincident, the RDcost is higher than that when configuring the PU partition block such as a of FIG. 6. It is expected to have a lower value.

Therefore, according to the method of Equation 1, RDcost for each of the eight candidate PU partition blocks of FIG. 5 may be calculated. In this step, when RDcosts of all candidate PU partition blocks available for inter prediction are calculated in each of the two selected basic unit blocks, the PU prediction execution unit 240 then performs the PU prediction according to a predetermined rule. (See step S340 of FIG. 3).

In this case, the predetermined rule applied in step S340 of FIG. 3 may be as follows.

That is, in the inter prediction process of the unit block having the lower skip cost among the two unit blocks selected in the previous step S320, the first number of PUs predetermined in the order of low RDcost for the PU partition block calculated in the previous step S330 PU prediction may be performed using a partition block. In addition, in the inter prediction process of the unit block having a higher skip cost among the two selected unit blocks, a second number previously specified in the order of low RDcost for the PU partition block in step S330 (here, the second number) PU prediction may be performed using a PU partition block of a number smaller than the first number.

The above rule is described with reference to the example of a of FIG. 4 as follows. Since a of FIG. 4 corresponds to a case where the 64x64 block has the lowest skip cost, for this 64x64 CTB block, among the total eight candidate PU partition blocks as shown in FIG. 5, the RDcost calculated in the previous step S330 has a low value. PU prediction may be performed by selecting a first predetermined number of PU partition blocks (eg, five) in order and using the same. In the case of FIG. 4, the 32x32 block has the lowest skip cost after the 64x64 block. For this 32x32 CTB block, the lower RDcost value calculated in step S330 among the eight candidate PU partition blocks shown in FIG. 5 is the lower value. PU prediction blocks may be selected by using a second predetermined number (eg, three) of PU partition blocks in the order of, and PU prediction may be performed using the PU partition blocks.

As described above, in this step, a method of performing PU prediction by allocating more candidate PU partition blocks for inter prediction of a unit block having a lower skip cost among the two selected basic unit blocks. To be adopted. In addition, in this step, in the PU prediction process of the selected two basic unit blocks, instead of using all candidate PU partition blocks, some of the predetermined number of PU partition blocks having a low RDcost is used to reduce the computational burden. Decrease.

An experimental example of this is shown in the table of FIG. 7. Referring to FIG. 7, the proposed PU pruning method according to an embodiment of the present invention reduces the computation time by approximately 11% compared to the conventional early coding unit (ECU) pruning method (that is, the ECU Runtime reduction of 37.49, the proposed method shows a 48.45% runtime reduction).

Although the above has been described with reference to embodiments of the present invention, those skilled in the art may variously modify the present invention without departing from the spirit and scope of the present invention as set forth in the claims below. And can be changed easily.

Claims

A computer-implemented method for a pruning technique of a prediction unit (PU) for high efficiency video coding (HEVC) inter prediction,

(a) calculating a skip cost for each of a 64x64 block, a 32x32 block, and a 16x16 block corresponding to a basic unit block of a coding tree block (CTU);

(b) selecting two unit blocks to perform PU prediction among the three unit blocks in order of having a relatively low skip cost through comparing the calculated skip costs; And

(c) perform PU prediction on the selected two unit blocks, and more candidate PU partition blocks for inter prediction of a unit block having a lower skip cost among the selected two unit blocks; Performing PU prediction by assigning

PU pruning method for HEVC inter prediction.
The method of claim 1,

Calculating, for each of the two selected unit blocks, a respective rate-distortion cost (RD cost) for all candidate PU partition blocks available for inter prediction. PU pruning method for prediction.
The method of claim 2,

In step (c),

For inter prediction of a unit block having a lower skip cost among the two selected unit blocks, PU prediction is performed using a first number of PU partition blocks that are predetermined in the order of low rate-distortion cost for the PU partition block. Performing; And

A second number pre-specified in order of low rate-distortion cost for a PU partition block for inter prediction of a unit block having a higher skip cost among the two selected unit blocks, wherein the second number is the first number; Performing PU prediction using a PU partition block having a number less than 1)

PU pruning method for HEVC inter prediction.
The method of claim 2,

The rate-distortion cost for the PU partition block is calculated by Equation 1 below.

[Equation 1]

puPairCost = ∑│puMvMedian-skipMv│ × skipRDCost

Here, puPairCost represents a rate-distortion cost for a specific PU partition block, skipMv represents a skip motion vector of each 8x8 block in the corresponding PU partition block, and puMvMedian is an intermediate of skipped motion vectors of all 8x8 blocks in the corresponding PU partition block. A median value, and skipRDCost represents a rate-distortion cost of a skip motion vector of each 8x8 block.
A device to which a pruning method of a prediction unit (PU) for high efficiency video coding (HEVC) inter prediction is applied,

A skip cost calculator configured to calculate a skip cost for each of 64x64 blocks, 32x32 blocks, and 16x16 blocks corresponding to basic unit blocks of a coding tree block (CTU);

A PU prediction determination unit selecting two unit blocks for performing PU prediction among the three unit blocks in an order having a relatively low skip cost by comparing the calculated skip costs; And

Perform PU prediction on the selected two unit blocks, and allocate more candidate PU partition blocks for inter prediction of a unit block having a lower skip cost among the selected two unit blocks. PU prediction execution unit that performs PU prediction

PU prediction determination apparatus for HEVC inter prediction comprising a.
The method of claim 1,

For each of the two selected unit blocks, HEVC further comprises an RDcost calculator for calculating each of the rate-distortion cost (RD cost: RD cost) for all candidate PU partition blocks available for inter prediction PU prediction determination apparatus for inter prediction.
The method of claim 2,

The PU prediction execution unit,

For inter prediction of a unit block having a lower skip cost among the two selected unit blocks, PU prediction is performed using a first number of PU partition blocks that are predetermined in the order of low rate-distortion cost for the PU partition block. Doing,

A second number pre-specified in order of low rate-distortion cost for a PU partition block for inter prediction of a unit block having a higher skip cost among the two selected unit blocks, wherein the second number is the first number; The PU prediction determining apparatus for HEVC inter prediction using a PU partition block of less than one number).
The method of claim 6,

An RDcost calculator calculates a rate-distortion cost for the PU partition block by using Equation 2 below.

[Equation 2]

puPairCost = ∑│puMvMedian-skipMv│ × skipRDCost

Here, puPairCost represents a rate-distortion cost for a specific PU partition block, skipMv represents a skip motion vector of each 8x8 block in the corresponding PU partition block, and puMvMedian is an intermediate of skipped motion vectors of all 8x8 blocks in the corresponding PU partition block. A median value, and skipRDCost represents a rate-distortion cost of a skip motion vector of each 8x8 block.