CN100551075C - A low-complexity intra-frame prediction mode selection method - Google Patents

Abstract

The invention discloses a kind of method for choosing frame inner forecast mode of low complex degree, in current block, get two groups of sample points that characterize textural characteristics, overhead value according to adjacent block, calculated threshold, this method may further comprise the steps: judge whether current block is in the specific position of picture frame, if, calculate optimal prediction modes and output under this specific position, finish to calculate; If not, calculate the overhead value of current block under the most probable predictive mode, if less than threshold value, then export this pattern, finish to calculate; Otherwise, select overhead value minimum in all non-most probable predictive modes one, as interim optimal prediction modes; Relatively the overhead value of most probable predictive mode and interim optimal prediction modes is selected wherein less exporting as optimal prediction modes.The present invention replaces whole data block by selected characteristic point, under the situation that video encoding quality is not had obviously influence, significantly reduces the amount of calculation that intra prediction mode is selected, and improves the real-time of video coding.

Description

A kind of method for choosing frame inner forecast mode of low complex degree

Technical field

The invention belongs to the video information compression field, be specifically related to a kind of method for choosing frame inner forecast mode of low complex degree.

Background technology

All comprised the infra-frame prediction function in the present advanced video coding standard, utilize the sample point of adjacent blocks to do extrapolation and realize prediction current block, with the spatial redundancy in the better elimination single-frame images, so only need encode to the residual error of prediction piece and current block.Especially changing smooth zone, utilize infra-frame prediction can reduce code check greatly.

When a macro block is when adopting the frame mode coding, utilize a prediction of the block structure piece P of the previous encoded structure of laying equal stress on.For luminance component, can be each piece or macro block establishment prediction piece P.For example H.264 in the coding standard 4 * 4 luminance block have 9 kinds of alternative modes, 16 * 16 luminance block have 4 kinds of alternative modes.

As shown in Figure 1, in standard H.264, utilize the several or all points in 13 sample points (A～L and Q) of having decoded in the adjacent block, predict sample point in current 4 * 4 luminance block (a～p).Select best a kind of of effect in 9 kinds of predictive modes, as the optimum prediction mode of this piece.9 kinds of predictive modes comprise: the mean prediction of pattern 2 (DC_PRED) and 8 kinds of direction predictions as shown in Figure 2.

Each predictive mode of 4 * 4 all encoded need take a lot of bits, and the suitable correlation of utilizing the space adjacent block can reach the purpose of high efficient coding.As shown in Figure 3, C is 4 * 4 current luminance block, obtains a most probable predictive mode of C piece according to the various combination of A piece and B block prediction mode.The most probable predictive mode that obtains is compared with the optimum prediction mode of aforementioned C piece, if it is identical, then only need when coding, to use 1 bit to represent most probable predictive mode, represent to remain one best in 8 kinds of predictive modes otherwise only need send 3 bits.

In traditional way, use all direction search method of 9 kinds of patterns to find optimum a kind of predictive mode, main step is as follows:

1, goes out 4 * 4 prediction piece P according to a kind of schema construction;

2, absolute error and the SAD between calculating original block and the prediction piece P ₁₆

3, computing cost value Cost ₁₆=SAD ₁₆+ 4R λ (QP) (1)

${\mathrm{SAD}}_{16}=\underset{x=0}{\overset{3}{\mathrm{\Σ}}}\underset{y=0}{\overset{3}{\mathrm{\Σ}}}|s(x,y)-s^{\′}(x,y)|---\left(2\right)$

Wherein, λ (QP) is the exponential function of quantizing factor QP, and QP is according to different standards, the value difference; λ (QP) is used for adjusting predictive mode in the shared ratio of output code flow, and QP is big more, and the predictive mode proportion is high more; S (x, y) (x y) represents original point and future position respectively, and x, y are its coordinate figure with s '; According to adjacent block current block is predicted, obtained the most probable predictive mode of current block, whether R is used for difference is the most probable predictive mode, R=0 when present mode is most probable predictive mode, R=1 under other 8 kinds of situations;

4, repeat the 1-3 step, from 9 kinds of prediction mode, select Cost ₁₆Be worth minimum one, be best predictive mode.

Though this all direction search method can find best predictive mode, its amount of calculation is very big, is an a lot of part consuming time in the frame and in the interframe encode.

Patent CN200410006340 disclosed method by calculating the textural characteristics of macroblock to be encoded, is selected optimum predictive mode according to the grain direction in the textural characteristics then.This method need be carried out the macroblock texture analysis, and as shade of gray method, fourier spectrum analytic approach etc., computation complexity is higher, is unfavorable for the pad optimization in system's later stage.

Patent CN200480006978 disclosed method by calculating the edge directional information of Intra-coded blocks, is carried out the selection of optimal prediction modes then.This method need be determined the amplitude and the angle of the edge vectors of all pixels in the piece, is each pixel edge calculation direction histogram, and amount of calculation is very big, and practicality is not high.

Summary of the invention

In view of this, it is a kind of practical that main purpose of the present invention is to provide, and calculates the method for choosing frame inner forecast mode of easy low complex degree.

For achieving the above object, technical scheme of the present invention is achieved in that

A kind of method for choosing frame inner forecast mode of low complex degree is got two groups of sample points that characterize textural characteristics in current block, according to the overhead value of adjacent block, and calculated threshold, this method may further comprise the steps:

A, judge whether current block is in the specific position of picture frame, if, calculate optimal prediction modes and output under this specific position, finish to calculate; If not, enter step B;

If B, the overhead value of calculating current block under the most probable predictive mode less than threshold value, are then exported this pattern, finish to calculate; Otherwise, enter step C;

C, select of overhead value minimum in all non-most probable predictive modes, as interim optimal prediction modes;

The overhead value of D, comparison most probable predictive mode and interim optimal prediction modes is selected wherein less exporting as optimal prediction modes.

Specific position is meant in the steps A, and current block is in upper left side, the top or the left of picture frame.

Optimal prediction modes under this specific position of the described calculating of steps A divides three kinds of situations:

Current block is positioned at the upper left side of picture frame, and optimal prediction modes is 2;

Current block is in the top of picture frame, calculates pattern 1, pattern 2 and pattern 8 times, and the overhead value of described two groups of sample points is selected the optimal prediction modes of the pattern of overhead value minimum as current block;

Current block calculates under pattern 0, pattern 2, mode 3 and mode 7 at the left of picture frame, and the overhead value of described two groups of sample points is selected the optimal prediction modes of the pattern of overhead value minimum as current block.

The step of calculating interim optimal prediction modes overhead value among the step C comprises:

C1, calculate first group of sample point under all non-most probable predictive modes absolute error and, absolute error and minimum pattern are designated as interim optimal prediction modes M ₁, M ₁The pattern of left is M ₂, M ₁Right-hand pattern is M ₃

Whether C2, judgment model 2 are most probable predictive mode, if, note M ₄Be sky, enter step C4; Otherwise note pattern 2 is M ₄, enter step C3;

C3, judgement M ₂Whether be most probable predictive mode, if remember M again ₂The pattern of left is M ₂, enter step C4; Otherwise, judge M ₃Whether be most probable predictive mode, if remember M again ₃Right-hand pattern is M ₃, enter step C4;

C4, difference computation schema M ₁, M ₂, M ₃And M ₄Second group of sample point absolute error and, and get absolute error and addition with corresponding first group of sample point, obtain each pattern absolute error and, select absolute error and minimum pattern as interim optimal prediction modes, enter step D.

According to the textural characteristics of image block, counting that two groups of sample points comprise is identical or different.

Described method is applicable to that smallest blocks is 4 * 4 picture frame.

The present invention replaces whole data block by selected characteristic point, under the situation that video encoding quality is not had obviously influence, significantly reduces the amount of calculation that intra prediction mode is selected, and improves the real-time of video coding.

Description of drawings

Fig. 1 is 4 * 4 forecast sample point schematic diagrames;

Fig. 2 is 8 kinds of prediction direction schematic diagrames of 4 * 4 predictions;

Fig. 3 is adjacent block A, B, C graph of a relation;

Fig. 4 is the flow chart of the method for the invention.

Embodiment

Main thought of the present invention is: the computing function of conforming to the principle of simplicity overhead value (Cost) and minimizing need pattern count two aspects of checking to start with, and intra prediction mode is selected to handle.The details that common one 4 * 4 luminance block has is less, and texture structure is smoother, so just can simplify Cost value computing function by calculative sample point number in minimizing 4 * 4 luminance block.In addition, in the infra-frame prediction based on 4 * 4 luminance block, optimum predictive mode has similar direction to the predictive mode of suboptimum usually; And when QP is very big (such as 40), penalty factor 4 λ (QP) in other patterns will increase, so most probable predictive mode is exactly optimum predictive mode usually; Especially, in the zone that picture material mixes very much, any predictive mode all can not reach good prediction effect.

With 4 * 4 shown in Figure 1 be example, the key step of the current block of this method (C piece) prediction is as follows:

Step 1: in current block, get two groups of sample points that characterize textural characteristics.

Get e, f, g, h, m, n, o and p in current 4 * 4 as first group of sample point, a, b, c, d, i, j, k and l are as second group of sample point.When calculating the Cost value, just adopt formula:

Cost _n＝SAD _n+Rλ(QP) (3)

Wherein, SAD _nBe to utilize the absolute error that calculates of n sample point and, R and λ (QP) in each group identical with implication in the equation (1).

Above-mentioned two groups of sample point Cost _N1And Cost _N2And, be designated as Cost _mSAD _N1And SAD _N2And, be designated as SAD _m

Step 2: judge whether current block is in the specific position of picture frame, if, calculate optimal prediction modes and output under this position, withdraw from calculating; If not, enter step 3.

The specific position of so-called picture frame is meant that current block is in the upper left side of picture frame, the top or left.Wherein, the most upper left predictive mode of picture frame is 2; If current block is in the top of picture frame,, calculate the Cost of above-mentioned two groups of sample points then pattern 1, pattern 2 and pattern 8 times _mValue is selected Cost _mBe worth the optimal prediction modes of a minimum pattern, and withdraw from calculating as current block; If current block then under pattern 0, pattern 2, mode 3 and mode 7, calculates the Cost of above-mentioned two groups of sample points at the left of picture frame _mValue is selected Cost _mBe worth the optimal prediction modes of a minimum pattern, and withdraw from calculating as current block.If current block is not the specific position that is in picture frame, then enter step 3.

Step 3: according to the Cost of adjacent block _mValue, calculated threshold T.

Its Cost all can be preserved after each piece calculating as shown in Figure 3 in the adjacent block position _mBe worth, therefrom read the Cost of A piece, B piece _mValue.According to the signatures to predict C piece of A piece, B piece, obtain the most probable predictive mode of C piece, write down this pattern.If most probable predictive mode is exactly an optimal prediction modes, then the C piece has similar texture structure to A piece or B piece, and promptly the prediction residual of the prediction residual of C piece and A piece or B piece should be very approaching.The effect of penalty factor λ (QP) is taken into account, T is set at threshold value again:

T＝min(Cost _mA，Cost _mB)+λ(QP) (4)

Wherein, Cost _MABe the Cost of A piece _mValue, Cost _MBBe the Cost of B piece _mValue, λ (QP) is identical with implication in the equation (1).

Step 4: calculate the Cost of C piece under most probable predictive mode _mValue is designated as Cost _MCAt this moment, R=0 is according to (1) formula, Cost _MC=SAD _mIf Cost _MCThe result less than threshold value T, so just, withdraw from calculating its optimal prediction modes as current 4 * 4 luminance block, with most probable predictive mode as final result output; Otherwise illustrate that most probable predictive mode can not predict the C piece as the good result of prediction A piece or B piece, just must in other 8 kinds of predictive modes, select, enter step 5.

Step 5: by calculating the SAD of first group of sample point under other 8 kinds of patterns _N1, select interim optimal prediction modes and left and right sides pattern thereof.

Except most probable predictive mode, other predictive modes are calculating Cost _nAll to be subjected to the identical restriction of penalty factor λ (QP) during value, so when comparing the prediction effect of other 8 kinds of predictive modes, can not consider the influence of penalty factor earlier.Because pattern 2 is irrelevant with directivity, need to consider separately, pattern and pattern 2, the most probable predictive mode that calculates comprehensively compared, obtain interim optimal prediction modes and left and right sides pattern thereof.Concrete steps are as follows:

Step 501: the SAD that calculates first group of sample point under all non-most probable predictive modes _N1, obtain SAD _N1The pattern that value is minimum is designated as interim optimal prediction modes M ₁M ₁The pattern of left is designated as M ₂, M ₁Right-hand pattern is designated as M ₃Relation is as shown in table 1 about various patterns.

Predictive mode	The left pattern	Right-hand pattern
			0	5	7
1	8	6
			2	0	1
3	7	8
			4	6	5
5	4	0
			6	1	4
7	0	3
			8	3	1

Table 1

Step 502: because pattern 2 is irrelevant with directivity, can't embody in the table 1,, use M so in calculating, consider separately ₄The relation of expression pattern 2 and most probable predictive mode, whether judgment model 2 is most probable predictive mode, if, note M ₄Be sky, enter step 6; Otherwise note pattern 2 is M ₄, enter step 503.

Step 503: judge M ₂Whether be most probable predictive mode, if remember M again ₂The pattern of left is M ₂, enter step 6; Otherwise, judge M ₃Whether be most probable predictive mode, if remember M again ₃Right-hand pattern is M ₃, enter step 6.

Step 6: calculate the predictive mode M that in step 5, obtains respectively ₁, M ₂, M ₃And M ₄The SAD of second group of sample point _N2Be worth, and get SAD with its first group of sample point separately _N1SAD is selected in the value addition _mBe worth the minimum new interim optimal prediction modes of a conduct, calculate its Cost _mValue is designated as Cost _MM, enter step 7.

Step 7: compare Cost _MCAnd Cost _MM, select wherein less one to export as optimal prediction modes.If the two equates that variation has taken place the direction details of key diagram picture, this moment, the prediction effect of interim optimal prediction modes was better than most probable predictive mode.At this moment select interim optimal prediction modes as final optimal prediction modes, have better prediction effect for the follow-up right-hand and piece below so.

In order to find optimal prediction modes, each 4 * 4 needs calculate the Cost values of 16 * 9=144 sample point under all direction search method.And the best situation of the method for fast searching among the embodiment is only to need calculation procedure 1～4, the Cost value of following 16 points of promptly most probable predictive mode.Find that through experiment when the QP value increased, the probability that only calculates the most probable predictive mode also increased thereupon.For example be about 30% when QP=16, increasing during QP=31 is 50%, and maximum is near 80% during QP=48.

The poorest situation is to need computation schema 2 in the step 5, and the sample point of Cost value adds up to:

(16*1)+(8*8)+(8*4)＝112

Even under the poorest situation, the quick mode searching method among the embodiment has still been saved the amount of calculation of 32 sample points as can be seen, the time efficiency that makes predictive mode select improves about 20%.

In the step 1, counting that two groups of sample points comprise can be different, according to the textural characteristics of image block, choose the point that can characterize textural characteristics, as sample point.In the step 3 threshold value T choose most important: if T is less, then can not effectively reduce calculative predictive mode number; Otherwise, be easier to select most probable predictive mode as final result, can not effectively find the optimum prediction mode of current block.General, in the zone of picture material more complicated, the value of increase T that can be suitable; Otherwise, should reduce the value of T.In addition, the size of the selection of T and QP value also has relation: when QP was big, predicated error was bigger, the value of increase T that then should be suitable; Otherwise, should reduce the value of T.

The above is preferred embodiment of the present invention only, is not to be used to limit protection scope of the present invention.

Claims (5)

1. A low-complexity intra-frame prediction mode selection method, characterized in that, in the current block, two groups of sample points representing texture features are taken, and the threshold is calculated according to the overhead value of the adjacent block, the method comprising the following steps: A. Determine whether the current block is in a special position of the image frame, if yes, calculate and output the optimal prediction mode at the special position, and end the calculation; if not, enter step B; B. Calculate the overhead value of the current block in the most probable prediction mode, if it is less than the threshold, then output the mode and end the calculation; otherwise, enter step C; C. Select the one with the smallest cost value among all non-most probable prediction modes as the temporary optimal prediction mode; D. Compare the overhead values of the most probable prediction mode and the temporary optimal prediction mode, and select the smaller one as the output of the optimal prediction mode; The step of calculating the temporary optimal prediction mode overhead value in the step C includes: C1, calculating the absolute error sum of the first group of sample points under all non-most probable prediction modes, and the mode with the smallest absolute error sum is recorded as the temporary optimal Prediction mode M ₁ , the mode on the left of M ₁ is M ₂ , and the mode on the right of M ₁ is M ₃ ; C2, judge whether mode 2 is the most probable prediction mode, if yes, record M ₄ as empty, and go to step C4 ;Otherwise, record mode 2 as M ₄ , and enter step C3; C3, determine whether M ₂ is the most probable prediction mode, if so, re-record the mode to the left of M ₂ as M ₂ , and enter step C4; otherwise, determine M ₃ Whether it is the most probable prediction mode, if so, re-record the mode on the right of M ₃ as M ₃ , and go to step C4; C4, respectively calculate the second group of sample points of modes M ₁ , M ₂ , M ₃ and M ₄ The sum of absolute errors and the sum of absolute errors of the corresponding first group of sample points are added to obtain the sum of absolute errors of each mode, and the mode with the smallest sum of absolute errors is selected as the temporary optimal prediction mode, and enters step D. 2. The method for selecting a low-complexity intra-frame prediction mode according to claim 1, wherein the special position in step A means that the current block is at the uppermost left, uppermost or leftmost of the image frame. 3. The low-complexity intra-frame prediction mode selection method according to claim 2, wherein the calculation of the optimal prediction mode at the special position in step A is divided into three situations: The current block is located at the top left of the image frame, and the optimal prediction mode is 2; The current block is at the top of the image frame, and the cost values of the two groups of sample points are calculated under mode 1, mode 2 and mode 8, and the mode with the smallest cost value is selected as the optimal prediction mode of the current block; The current block is at the far left of the image frame, and the cost values of the two groups of sample points are calculated in mode 0, mode 2, mode 3 and mode 7, and the mode with the smallest cost value is selected as the optimal prediction mode of the current block. 4. The low-complexity intra-frame prediction mode selection method according to claim 1, characterized in that, according to the texture feature of the image block, the numbers of points included in the two groups of sample points are the same or different. 5. The method for selecting a low-complexity intra-frame prediction mode according to claim 1, wherein the method is applicable to an image frame whose minimum block is 4×4.

Images