CN103517069B - A kind of HEVC intra-frame prediction quick mode selection method based on texture analysis - Google Patents

Abstract

The invention discloses a kind of HEVC intra-frame prediction quick mode selection method based on texture analysis, the method is before carrying out infra-frame prediction to code tree unit, according to the gradient absolute value on the four directions such as level, vertical, lower-left, bottom right and the main grain direction and the Texture complication that determine each Unit 4 × 4 in code tree unit, and using large code unit according to texture smooth region, texture complex region uses the principle of relatively lower Item unit to determine the division of present encoding tree unit.When prediction, according to the main grain direction of predicting unit, exclude some predictive modes most unlikely, then according to HEVC coding standard carries out coarse mode selection and R-D optimized mode selection.HEVC intra-frame prediction quick mode selection method based on texture analysis proposed by the invention can significantly improve coding rate on the premise of ensureing coding quality.

Description

A kind of HEVC intra-frame prediction quick mode selection method based on texture analysis

Technical field

The invention belongs to multimedia coding field, be specifically related to a kind of infra-frame prediction Fast Modular for HEVC video encoding standard Formula system of selection, is a kind of method carrying out coding unit division and prediction direction fast mode decision according to analyzing image texture.

Background technology

Along with a new generation video encoding standard HEVC(High Efficiency Video Coding) issue, video coding effect Rate promotes further.Compared to former video encoding standard H.264/AVC, in the case of identical coding quality, Ke Yijie Save the code check of nearly half, but this is to sacrifice encoder complexity as cost.How not change code flow structure and to keep original On the premise of code check, farthest reduce algorithm complex, significantly improve coding rate, be HEVC large-scale promotion with should Key point.

Intraframe coding can be that interframe encode provides reference information, and can stop the error drift caused by error code, therefore exists In Video coding, intraframe coding is requisite.Affect HEVC intraframe coding time-consuming mainly have two parts: coding unit Division and the selection of prediction direction.

Quick division for coding unit mainly has two kinds of methods, and one is that the rate distortion costs according to encoded unit utilizes pattra leaves The method of this statistical learning such as grade determines the division of coding unit, but the learning outcome of off-line can not well be common to all of regarding Frequency sequence, online learning style can introduce again extra computation complexity；Two is that the picture material according to current coded unit is multiple Miscellaneous degree determines the division of coding unit, and it is critical only that of this method calculates image complexity and select suitable threshold value.

Fast mode decision for prediction direction mainly has three kinds of methods, and one is to have relatively according to adjacent area or adjacent interframe Strong dependency, utilizes the information of coding unit around to reduce the prediction direction Candidate Set of current coded unit, but reality is applied In, scene switching or the video of texture-rich often occur, temporal correlation is more weak in this case, causes coding efficiency obvious Decline；Two is to utilize dependency higher between the grain direction of optimal prediction direction and coding unit, reduces candidate prediction direction Quantity；Three is code check and the distortion reduction rate distortion cost estimating each predictive mode by the way of a kind of low complex degree Computation complexity, this method remains all of prediction direction, and does not increase extra computation complexity, but due to The randomness of natural image video sequence is relatively strong, is difficult to one rate distortion estimation model being applicable to all video sequences of structure, phase Ratio is in second method, and the coding efficiency suppression ratio of this method is more apparent.

In a new generation's video encoding standard, it is general that HEVC introduces new code tree unit (Coding Tree Unit, CTU) Read, for substituting the macro block concept in existing video encoding standard, the most also coding unit, predicting unit, converter unit etc. Concept.Code tree unit is the quad-tree structure of a recurrence, and the degree of depth can be from 0 to 3, i.e. the size of coding unit can be from 64 × 64 to 8 × 8.For minimum coding unit, the intraframe coding of HEVC can be divided into again 4 × 4 pre- Survey unit, it was predicted that the size of unit can be from 64 × 64 to 4 × 4.And for infra-frame prediction, HEVC is pre-to each Survey unit and both provide 35 kinds of predictive modes.Compared to existing video encoding standard H.264/AVC, coding unit divides and pre- The optimization problem surveying set direction is increasingly complex.The present invention retouches according to the image complexity of texture analysis structure present encoding tree unit State, and the most quickly determine the division of coding unit；Utilize phase stronger between grain direction that texture analysis obtains with prediction direction Guan Xing, reduces the quantity of predictive mode.

Summary of the invention

In order to overcome the drawbacks described above of prior art, it is quick that the present invention proposes a kind of HEVC infra-frame prediction based on texture analysis Mode selecting method, the method determines the division of code tree unit according to the complexity of video content, and according to the texture side of image To the Candidate Set reducing prediction direction.Experiment proves that this method can be good at adapting to the Real Time Compression application of high clear video image, Substantially increase coding rate.

For achieving the above object, the technical solution used in the present invention is:

A kind of HEVC intra-frame prediction quick mode selection method based on texture analysis, the method step is as follows:

Step (1), to present encoding tree unit, calculate the image complexity of each possible coding unit and main grain direction；

Step (2), to present encoding tree unit, according to Z scanning sequency repeated execution of steps from top to bottom (3) to step (6), Until scanning through present encoding tree unit；

Step (3) if the image complexity of current coded unit is less than or equal to threshold value, then judges that this coding unit is as smooth single Unit, is otherwise judged to complex unit；

Step (4), for smooth coding unit, utilize step (5) to calculate the rate distortion costs of current coded unit；

Step (5), for each predicting unit in current coded unit, retain 8 around the main grain direction of this unit Prediction direction and planar predict and DC predictive mode, carry out coarse mode selection and rate according to HEVC video encoding standard Aberration optimizing selects；

Step (6), for complicated coding unit, according to step (3) to four subelements of step (6) successively recursive scanning, If four subelements are all judged to smooth unit, then step (5) is utilized to calculate the rate distortion costs of current coded unit, and root Judge whether current coded unit divides, and updates threshold value simultaneously according to rate distortion costs minimum principle.

Described step (1) specifically includes following steps:

Step (11), to each unit of 4 × 4 in present encoding tree unit, respectively along level, vertical, lower-left, the right side Lower four direction calculates gradient and calculates absolute value and the SAG of gradient；

Step (12), using the gradient direction of minimal gradient absolute value sum as the main grain direction of this unit, described main texture side To the grain direction referring to this unit major part pixel；

Step (13), the absolute value of SAG difference on orthogonal to that for main grain direction direction is defined as the Texture complication of this unit, And utilize the quantization step product with this cell size as initial threshold.

Described step (5) specifically includes following steps:

Step (51), one candidate modes array of definition；

Step (52), obtain the main grain direction of current prediction unit according to step (1), and by main grain direction and about 8 prediction direction as candidate prediction direction；

Step (53), the angle predictive mode that candidate prediction direction is mapped in HEVC coding standard, and the angle that will select Degree predictive mode and planar and DC pattern add candidate modes array to；

Step (54), to each predictive mode in candidate modes array, carry out roughly according to HEVC coding standard Model selection and R-D optimized mode selection.

Described step (6) specifically includes following steps:

Step (61), according to step (3) to four subelements of step (6) successively recursive scanning, if four subelements are all sentenced It is set to smooth unit, then utilizes step (5) to calculate rate distortion costs Cost of current coded unit_coarse, otherwise continue recurrence Scan the subelement of next level；

Step (62) if rate distortion costs Cost of current coded unit_coarseRate distortion generation less than or equal to four subelements Valency and Cost_fine, then threshold value is tuned up, otherwise threshold value is turned down.

The more New Policy that in described step (62), threshold value tunes up is:

TH '=TH+ α × (Cost_fine-Cost_coarse)/2

In formula, TH is the threshold value before updating, and TH ' is the threshold value after updating, and α is the line between Texture complication and rate distortion costs The multiplication factor of property model.

The more New Policy that in described step (62), threshold value is turned down is:

TH '=TH-a

In formula, TH is the threshold value before updating, and TH ' is the threshold value after updating, and a is constant.

Present invention advantage compared with prior art is:

1, the present invention is from the angle of texture analysis, the division of code tree unit is taken into consideration with the selection of prediction direction, Utilize gradient can either extract Texture complication and can extract again the direction of texture, on the premise of ensureing coding quality, Ke Yijie Save the more scramble time.

2, the present invention upgrades in time according to final rate distortion costs and determines the threshold value whether coding unit divides, it is possible to fully profit With the strong correlation of adjacent area in code tree unit, choose the impact on coding quality weakening threshold value.

Accompanying drawing explanation

Fig. 1 is HEVC intra-frame prediction quick mode selection method flow chart based on texture analysis；

Fig. 2 is the code tree dividing elements structure chart in HEVC video encoding standard；

Fig. 3 is direction gradient computational methods schematic diagram；

Fig. 4 is that coding unit based on gradient quickly divides detail flowchart.

Detailed description of the invention

The present invention is further illustrated with detailed description of the invention below in conjunction with the accompanying drawings.

In order to meet the Real Time Compression application demand of high clear video image, improve the coding rate of HEVC video encoder, this Bright propose a kind of HEVC intra-frame prediction quick mode selection method based on texture analysis.The method mainly includes two parts: Coding unit based on Texture complication quickly divides and predictive mode based on grain direction quickly selects, coding unit quick Divide the step (1) relating in summary of the invention to step (4) and step (6), it was predicted that the quickly selection of pattern relates to invention Step (5) in content.Calculate the complexity of each subelement in present encoding tree unit first with gradient operator, determine one Thinner coding unit divides, and determines that a thicker coding unit divides, finally according to follow-up infra-frame prediction further according to this division Rate distortion costs determine that final coding unit divides；During selecting optimum prediction direction, exclude and main texture side To difference predictive mode farther out, to save the time of infra-frame prediction.Developed below illustrate.

1. coding unit based on Texture complication quickly divides

A new generation's video encoding standard HEVC introduces code tree unit (Coding Tree Unit, CTU) concept, a frame video It is made up of the code tree unit of multiple non-overlapping copies.Under default situations, the size of code tree unit is 64 × 64.Code tree is single Unit is the recursive structure of a quaternary tree form, and it can be divided into the coding unit of many levels flexibly, as shown in Figure 1. The size of coding unit can be from 64 × 64 to 8 × 8.In the reference encoder device of HEVC standard, utilization rate aberration optimizing selects The division that each is possible is the most once predicted, converts, is quantified by the method selected, entropy code, inverse quantization, inverse transformation, weight The process such as building, compared to H.264/AVC, encoder complexity is substantially improved, and seriously constrains the actual application of HEVC.

Being found by experiment statistics, the size of coding unit is closely related with the Texture complication of picture material.The simple district of texture Territory uses bigger coding unit, and the region of texture complexity uses less coding unit.The metric form of Texture complication has Multiple, such as variance, gradient, entropy etc..Gradient not only comprises the Texture complication information of image, further comprises the texture side of image To information, can assist quickly to determine the predictive mode of predicting unit.Based on this, the present invention proposes a kind of coding based on gradient The quick division methods of unit.

Before introducing the embodiment that coding unit quickly divides, first provide the computational methods of gradient used in the present invention And main grain direction, the definition of Texture complication.Gradient in the present invention refer to pixel brightness value along level, vertical, lower-left, The directional derivative that the directions such as bottom right obtain, as shown in Figure 3.Computing formula is as shown in (1) (4) formula.

$\left\{\begin{array}{c}{g}_h(x,y)=p(x,y)-p(x+2,y)\\ g_h(x+1,y)=p(x+1,y)-p(x+3,y)\\ g_h(x,y+1)=p(x,y+2)-p(x+2,y+2)\\ g_h(x+1,y+1)=p(x+1,y+2)-p(x+3,y+2)\end{array}\right.---\left(1\right)$

$\left\{\begin{array}{c}{g}_v(x,y)=p(x,y)-p(x,y+2)\\ g_v(x+1,y)=p(x+2,y)-p(x+2,y+2)\\ g_v(x,y+1)=p(x+1,y+1)-p(x+1,y+3)\\ g_v(x+1,y+1)=p(x+3,y+1)-p(x+3,y+3)\end{array}\right.---\left(2\right)$

$\left\{\begin{array}{c}{g}_{\mathrm{ld}}(x,y)=\sqrt{2}[p(x,y)-p(x+2,y+2)]\\ g_{\mathrm{ld}}(x+1,y)=\sqrt{2}[p(x+1,y)-p(x+3,y+2)]\\ g_{\mathrm{ld}}(x,y+1)=\sqrt{2}[p(x,y+1)-p(x+2,y+3)]\\ g_{\mathrm{ld}}(x+1,y+1)=\sqrt{2}[p(x+1,y+1)-p(x+3,y+3)]\end{array}\right.---\left(3\right)$

$\left\{\begin{array}{c}{g}_{\mathrm{rd}}(x,y)=\sqrt{2}[p(x+2,y)-p(x,y+2)]\\ g_{\mathrm{rd}}(x+1,y)=\sqrt{2}[p(x+3,y)-p(x+1,y+2)]\\ g_{\mathrm{rd}}(x,y+1)=\sqrt{2}[p(x+2,y+1)-p(x,y+3)]\\ g_{\mathrm{rd}}(x+1,y+1)=\sqrt{2}[p(x+3,y+1)-p(x+1,y+3)]\end{array}\right.---\left(4\right)$

G in formula_h(·)、g_v(·)、g_ld(·)、g_rd() represents the gradient filtering in level, vertical, lower-left, lower right respectively Device, p () represents the luminance component pixel value in present encoding tree unit, and (x y) represents each 4 × 4 unit top left corner pixel Coordinate.The present invention utilizes SAG represent along a direction gradient absolute value and, computing formula is as shown in (5) formula.

SAG_dir=∑ | g_dir(x, y) |, dir ∈ { h, v, ld, rd} (5)

Using the direction of minimal gradient absolute value sum as main grain direction, the gradient absolute value of main grain direction and be denoted as SAG_min, The gradient absolute value of its orthogonal direction and be denoted as SAG_org.Texture complication complexity is defined as SAG_orgWith SAG_minBetween Difference, it may be assumed that

Complexity=SAG_org-SAG_min(6)

The detailed process divided is as shown in Figure 4.To each code tree unit, first it is divided into the list of N × N number of 4 × 4 Unit, to the Grad in each unit respectively calculated level, vertical, lower-left, lower right.Then to code tree unit from The Texture complication calculating each coding unit of 64 × 64 to 8 × 8 recurrence and division, finally determine whole code tree unit Divide.Recursive procedure is described as follows:

(1) to current coded unit, (5) formula of utilization calculates the gradient absolute value on four direction and SAG respectively_h、SAG_v、 SAG_ld、SAG_rd, and obtain main grain direction and the Texture complication complexity of current coded unit.

(2) if the Texture complication complexity of current coded unit is less than threshold value TH, directly current coded unit is arranged For smooth region, and calculation rate distortion cost, no longer divide, return last layer recurrence.

(3) otherwise, check whether current coded unit can divide, if cannot divide, then calculation rate distortion cost returning Return last layer recurrence.

(4) otherwise, current coded unit is divided into four subelements, sequentially enters next layer of recurrence.

(5) checking that four subelements are the most all confirmed as smooth region, smooth region then calculates current coded unit in this way Rate distortion costs, otherwise goes to step (8).

(6) if the rate distortion costs of current coded unit less than or equal to four subelements rate distortion costs and, the most currently compile Code unit does not divides, and threshold value TH is tuned up simultaneously.

(7) if the rate distortion costs of current coded unit more than four subelements rate distortion costs and, then current coded unit Divide, simultaneously by little for the fine setting of threshold value TH.

(8) detected whether uncoded unit, if having, having returned last layer recurrence, otherwise exiting recurrence.

It is, in general, that what the distortion that the encoder corresponding to existing video encoding standard produces mainly was caused by quantization, if adjacent Difference between pixel is less than quantization step, then illustrate that this difference does not exists after quantifying.Therefore can be walked by quantization The long threshold value setting Texture complication, but the quantization step in video encoding standard is for transform domain.The present invention is according to sky Law of conservation of energy between territory and transform domain, is defined as size and the quantization of current coded unit by the initial threshold in recursive procedure The product of step-length, i.e. TH=CUsize × QPstep.

The more New Policy that threshold value TH tunes up is: TH '=TH+ α × (Cost_fine-Cost_coarse)/2, α here is texture The multiplication factor of the linear model between complexity and rate distortion costs.

Threshold value TH finely tune little more New Policy be: TH ' be TH '=TH-a, a here be constant.

The division of code tree unit thus can be obtained according to Texture complication and rate distortion costs.

2. predictive mode based on grain direction quickly selects

Compared to H.264/AVC, a new generation's video encoding standard HEVC introduces predicting unit (Prediction Unit, PU) Concept.For infra-frame prediction, it was predicted that unit has two kinds of dividing mode.For minimum coding unit, it was predicted that unit is permissible Identical with coding unit, it is also possible to further four forks point on the basis of coding unit；And for remaining coding unit, it was predicted that single Unit can only be identical with coding unit.HEVC additionally provides the finest prediction direction simultaneously, and each predicting unit has 35 kinds of predictive modes, considerably increase encoder complexity.

Improve code efficiency be at all ensureing on the premise of coding quality, reduce candidate modes as far as possible.Normal conditions Under, the final predictive mode selected is close with grain direction.Table 1 gives optimum prediction mode and texture in various predicting unit The probability that direction is close.As can be seen from the table, the probability that optimum prediction mode is close with grain direction can reach more than 80%. It addition, in the quickly division of the first step coding unit of the present invention, have been obtained for the main grain direction of each coding unit. Therefore, it can the quantity utilizing this intermediate object program as far as possible to reduce candidate modes, thus improve code efficiency.Need explanation , for all of predicting unit, planar and the DC predictive mode that HEVC is provided all retains, and the present invention only considers Reduce the candidate prediction direction of angle prediction.

The each predicting unit of table 1 probability that optimum prediction mode is close with grain direction under different quantization steps

Quantization step	64×64	32×32	16×16	8×8	4×4
						22	100%	79.63%	80.76%	84.32%	87.45%
27	86.36%	78.80%	83.64%	87.67%	89.24%
						32	80.43%	82.21%	89.47%	88.88%	89.86%
37	78.67%	86.53%	89.83%	89.20%	89.06%
						Averagely	86.37%	81.79%	85.93%	87.52%	88.90%

The probability close due to optimum prediction mode and grain direction is higher, and the present invention utilizes the several predictions around main grain direction Direction carries out angle prediction.Detailed description of the invention is as follows:

(1) one size of definition is the candidate modes array of 11, for storing the candidate modes of screening, wherein wraps Include the some prediction direction around planar pattern, DC pattern, main grain direction and main grain direction.Tool in the present invention During body is implemented, 4 prediction direction are respectively taken in main grain direction both sides.

(2) prediction mode number mode corresponding to main grain direction of current prediction unit is obtained_texture, and utilize formula (7) Calculate angle estimation range [mode_left,mode_right]。

$\left\{\begin{array}{c}{\mathrm{mode}}_{\mathrm{left}}={\mathrm{mode}}_{\mathrm{texture}}-4\\ {\mathrm{mode}}_{\mathrm{right}}={\mathrm{mode}}_{\mathrm{texture}}+4\end{array}\right.---\left(7\right)$

(3) formula (8) is utilized to obtain the predictive mode on the left of main grain direction.

$\left\{\begin{array}{cc}\mathrm{mode}\mathrm{Idx}=\mathrm{mode}+33& \mathrm{mode}<2\\ \mathrm{mode}\mathrm{Idx}=\mathrm{mode}& \mathrm{mode}\&GreaterEqual;2\end{array}\right.,\mathrm{mode}\&Element;[{\mathrm{mode}}_{\mathrm{left}},{\mathrm{mode}}_{\mathrm{texture}}-1]---\left(8\right)$

(4) formula (9) is utilized to obtain the predictive mode on the right side of main grain direction.

$\left\{\begin{array}{cc}\mathrm{mode}\mathrm{Idx}=\mathrm{mode}-33& \mathrm{mode}>34\\ \mathrm{mode}\mathrm{Idx}=\mathrm{mode}& \mathrm{mode}\&le;34\end{array}\right.,\mathrm{mode}\&Element;[{\mathrm{mode}}_{\mathrm{texture}}+1,{\mathrm{mode}}_{\mathrm{right}}]---\left(9\right)$

(5) to the candidate modes set screened, { modeIdx} stores in candidate modes array.According to HEVC Coarse mode in standard selects and rate-distortion optimization selects therefrom to choose optimal predictive mode.

The specific embodiment being only the present invention disclosed above.The technological thought provided according to the present invention, those skilled in the art Can think and change, all should fall within the scope of protection of the present invention.

Claims (4)

1. a HEVC intra-frame prediction quick mode selection method based on texture analysis, it is characterised in that: the method step As follows:

Step (1), to present encoding tree unit, calculate the image complexity of each possible coding unit and main grain direction；

Step (2), to present encoding tree unit, according to Z scanning sequency repeated execution of steps from top to bottom (3) to step (6), Until scanning through present encoding tree unit；

Step (3) if the image complexity of current coded unit is less than or equal to threshold value, then judges that this coding unit is as smooth single Unit, is otherwise judged to complex unit；

Step (4), for smooth coding unit, utilize step (5) to calculate the rate distortion costs of current coded unit；

Step (5), for each predicting unit in current coded unit, only retain 8 around the main grain direction of this unit Individual prediction direction and planar prediction and DC predictive mode, according to HEVC video encoding standard carry out coarse mode select and Rate-distortion optimization selects；

Step (6), for complicated coding unit, according to step (3) to four subelements of step (6) successively recursive scanning, If four subelements are all judged to smooth unit, then step (5) is utilized to calculate the rate distortion costs of current coded unit, and root Judge whether current coded unit divides, and updates threshold value simultaneously according to rate distortion costs minimum principle；

Described step (6) specifically includes following steps:

Step (61), according to step (3) to four subelements of step (6) successively recursive scanning, if four subelements are all sentenced It is set to smooth unit, then utilizes step (5) to calculate rate distortion costs Cost of current coded unit_coarse, otherwise continue recurrence Scan the subelement of next level；

Step (62) if rate distortion costs Cost of current coded unit_coarseRate distortion generation less than or equal to four subelements Valency and Cost_fine, then threshold value is tuned up, otherwise threshold value is turned down；

The more New Policy that in described step (62), threshold value tunes up is:

TH '=TH+ α × (Cost_fine-Cost_coarse)/2

In formula, TH is the threshold value before updating, and TH ' is the threshold value after updating, and α is the line between Texture complication and rate distortion costs The multiplication factor of property model.

2. HEVC intra-frame prediction quick mode selection method based on texture analysis as claimed in claim 1, its feature exists Following steps are specifically included in: described step (1):

Step (11), to each unit of 4 × 4 in present encoding tree unit, respectively along level, vertical, lower-left, the right side Lower four direction calculates gradient and calculates absolute value and the SAG of gradient；

Step (12), using the gradient direction of minimal gradient absolute value sum as the main grain direction of this code tree unit, described master Grain direction refers to the grain direction of this code tree unit major part pixel；

Step (13), the absolute value of SAG difference on orthogonal to that for main grain direction direction is defined as the Texture complication of this unit, And utilize the quantization step product with this cell size as initial threshold.

3. HEVC intra-frame prediction quick mode selection method based on texture analysis as claimed in claim 1, its feature exists Following steps are specifically included in: described step (5):

Step (51), one candidate modes array of definition；

Step (52), obtain the main grain direction of current prediction unit according to step (1), and by main grain direction and about 8 prediction direction as candidate prediction direction；

Step (53), the angle predictive mode that candidate prediction direction is mapped in HEVC coding standard, and the angle that will select Degree predictive mode and planar and DC pattern add candidate modes array to；

Step (54), to each predictive mode in candidate modes array, carry out roughly according to HEVC coding standard Model selection and R-D optimized mode selection.

4. HEVC intra-frame prediction quick mode selection method based on texture analysis as claimed in claim 1, its feature exists In: the more New Policy that in described step (62), threshold value is turned down is:

TH '=TH-a

In formula, TH is the threshold value before updating, and TH ' is the threshold value after updating, and a is constant.