CN1756364A - Method for selecting intra-prediction mode - Google Patents

Abstract

The invention discloses a selection method for prediction mode in frame, which comprises steps: selecting reference image block for current being predicted image block of image frame; controlling the set of optimized prediction subschema, and selecting opposite optimized subschema from the set for in-frame prediction. This method can increase selection efficiency in H.264 compression coding process, and improve multimedia communication performance.

Description

The system of selection of intra prediction mode

Technical field

The present invention relates to the video compression coding-decoding technology in the multimedia communications system, relate in particular to a kind of system of selection of intra prediction mode.

Background technology

H.264 the compressed encoding standard is present state-of-the-art video compression coding international standard, its full name is that H.264 International Telecommunications Union-telecommunication standardization is organized suggestion (ITU-T is Recommendation H.264, International Telecommunication Union-Telecommunication is H.264Recommendation), below abbreviate it as H.264 standard.H.264 standard is by ITU-T and the (ISO of International Standards Organization, International Organization for Standardization)/(IEC of International Electrotechnical Commission, International Electrotechnical Commission) Motion Picture Experts Group (MPEG, Moving Picture Experts Group) coacts to develop and formulates.This standard is since development in 1999, by the equipment vendors of numerous countries, software developer, telecom operators, university and scientific research institution participate in, and have compiled state-of-the-art technology of video compressing encoding on the our times, formally become the international standard of ITU-T in 2003 the most finally.

H.264 with respect to formerly H.263+, H.263++ and video compression coding standard technique such as MPEG-4 Simple Profile, large increase has all been arranged on performance and function.Wherein on function, increased many coding tools options, made to be fit to more applications type and more wide application; At aspect of performance, under identical encoder bit rate (bit-rate), picture quality is with peak signal-noise ratio (PSNR, Peak Signal-to-Noise Ratio, be called for short Y-PSNR) as a reference, H.264 formerly H.263+/H.263++ standard compares, and MPEG-4 Simple Profile standard is doubled, and promptly PSNR has increased 3dB; That is to say that obtaining under the situation of identical PSNR, H.642 H.263+/H.263++ the needed network bandwidth of standard is compared, the required network bandwidth of MPEG-4 Simple Profile standard has reduced by 50%.

But, H.264 standard with respect to technical standard formerly when function and aspect of performance all are significantly improved, its cost of paying also is high, wherein H.264 standard increases several times to the computational complexity that video image carries out the compression coding and decoding processing than the computational complexity that H.263/H.263+ waits standard, therefore, H.264 the prerequisite of the extensive popularization and application of standard must be the higher arithmetic processing method of the multiple treatment effeciency of exploitation, to reduce the computational complexity of the very high calculating link of a plurality of calculating strength in the standard encoding and decoding calculation process process H.264, could reduce the disposal ability of the required multimedia processor of product like this, thereby reduce cost; Or multimedia processor supports higher coding and decoding capability on given disposal ability, thereby enhances product performance.

Wherein in the encoding and decoding processing procedure of standard H.264, there is the higher calculating link of many places computational complexity, such as inter prediction encoding (Inter-frame prediction, be called for short Inter), intraframe predictive coding (Intra-frame prediction, be called for short Intra), entropy coding (CABAC, Context-Adaptive BinaryArithmetic Coding), estimation (the ME of 1/4 pixel precision, Motion Estimation) and motion compensation (MC, Motion Compensation), multi-reference frame prediction (Multi-hypothesisprediction) or the like.

Especially intraframe predictive coding handle and the inter prediction encoding processing aspect, H.264 standard has had very big processing policy variation with respect to formerly the standard that H.263/H.263+ waits.Wherein H.264 standard for each macro block in the video image present frame (Macro block is hereinafter to be referred as MB), allows to use multiple predictive mode that it is carried out prediction processing in the process of video image being carried out the encoding and decoding processing.It at first is to carry out inter prediction that general forecast is handled, and then carry out infra-frame prediction, then the result who compares infra-frame prediction and inter prediction, choose the highest predictive mode of code efficiency, so just infra-frame prediction and these two kinds of Predicting Techniques of inter prediction have organically been combined, can reach higher forecasting efficiency and compression efficiency.

Wherein, infra-frame prediction is the exclusive processing attribute of standard H.264, and for the encoding video pictures process, infra-frame prediction can maximally utilise view data correlation spatially, thereby improves compression coding efficiency.In general video image compression coding processing mode, what comparatively generally adopt is inter prediction, and the inter prediction utilization is in time correlation between the sequence of video images; Yet its inside of video image of a lot of types is to have very strong spatial coherence, therefore H.264 standard design the infra-frame prediction mode making full use of the spatial coherence that video image inside exists, with so improve compression coding efficiency.

H.264 the infra-frame prediction in the standard is meant in cataloged procedure, for the current some zones that will encode in the video image (all being the piece of rectangle generally), search and its most akin and encoded piece in video image, predict the current piece that will encode with this most akin of searching, predicting the outcome promptly is the residual error of prediction.Wherein the close property between the different masses is to adopt absolute error and (SAD, Sum of Absolute Differences) to measure, i.e. the most akin of pairing conduct of minimum SAD that calculates with the current piece that will encode.Above-mentioned image block can be macro block MB (i.e. the piece of 16 * 16 pixel sizes), also can be the piece littler than macro block MB, that is to say that infra-frame prediction in the standard H.264 is more to carry out on the piece of small scale at MB or than MB.

For color video frequency image, a luminance component frame (YUV) and two chromatic component frames (Ycb and Ycr) are arranged respectively in every frame video image.Wherein the luminance component frame has two big intra prediction modes, and is as follows:

(1) Intra 16 * 16 intra prediction modes: be about to MB and make the as a whole infra-frame prediction that carries out.

(2) Intra 4 * 4 intra prediction modes: be about to each MB and be divided into 16 4 * 4 and carry out infra-frame prediction respectively, predicting the outcome of each piece pieced together get up to form 16 * 16 finally predicting the outcome then as this MB.

Then has only a kind of big intra prediction mode for two chromatic component frames, that is:

Intra 8 * 8 intra prediction modes promptly carry out infra-frame prediction with each 8 * 8 in the chromatic component frame as fundamental forecasting unit.

See also Fig. 1, this figure is existing H.264 compression coding technology carries out the sequential compression coding to each macro block MB a schematic diagram; Because H.264 compressed encoding is according to the putting in order from top to bottom of each MB, from left to right carry out successively, therefore for the current MB that will encode, the MB on above it and the left side has finished encoding process, is the MB that can be used as fundamentals of forecasting.Therefore just can utilize the current MB that will encode or 4 * 4 top delegation and pixel that the left side one is listed as to predict the value of each pixel among MB that this will be encoded or 4 * 4 as the reference pixel for the infra-frame prediction of luminance component frame.As shown in Figure 1, just can be when the pixel f among the MB that encode is carried out infra-frame prediction with reference to the top pixel a of delegation, b, c and d, and the left side one row pixel e carries out.

Because there are a plurality of directions in video image correlation spatially, process to simplify the process, H.264 standard code only on limited direction, carry out infra-frame prediction, see also Fig. 2, this figure is that existing H.264 standard is carried out the prediction direction schematic diagram that infra-frame prediction limited; As seen from Figure 2, when adopting Intra 4 * 4 intra prediction modes that the luminance component frame is carried out infra-frame prediction, the direction one that allows to use has 9 (each direction identifies with 0～8 numerical chracter in the drawings), wherein direction 2 is more special, be called " direct current " predictive mode, this predictive mode is to adopt all reference pixels to be carried out arithmetic average to handle and predict each pixel value in each 4 * 4, does not have particular orientation in essence, i.e. all directions equalization.

See also Fig. 3, this figure is the forecasting process schematic diagram that existing H.264 standard is carried out on each prediction direction shown in Figure 2 respectively; Wherein among Fig. 3 every from the dotted line of reference pixel the pixel of process all can utilize this reference image usually to carry out infra-frame prediction.Please consult Fig. 1 simultaneously, in direction 1 (level), from the dotted line of reference pixel I a, b, c and the d pixel of process all to utilize pixel I to carry out infra-frame prediction; And in the direction 5 (vertically taking over), from the dotted line of reference pixel A a, e, j and the n pixel of process all to utilize pixel A to carry out infra-frame prediction.This shows, when adopting Intra 4 * 4 intra prediction modes that the luminance component frame is carried out infra-frame prediction, have the prediction direction of 9 kinds of predictor patterns, 9 permissions corresponding shown in Figure 2 in fact.

And when adopting Intra 16 * 16 intra prediction modes that the luminance component frame is carried out infra-frame prediction, the predictor pattern that allows to use comprises 4: be respectively Mode0 (vertically) prediction, Mode1 (level) prediction, Mode2 (DC) prediction and Mode3 (plane) prediction; When adopting Intra 8 * 8 intra prediction modes that the colourity component frame is carried out infra-frame prediction, also allow to use 4 kinds of predictor patterns, be respectively: Mode0 (DC) prediction, Mode1 (level) prediction, Mode2 (vertically) prediction and Mode3 (plane) prediction.

According to prior art H.264 reference software realize technical scheme in (JM, Joint Mode1, JM have released a plurality of versions, at present highest version is JM 8.5), a complete intra prediction mode selection course is as follows:

For the current MB that will encode, at first be divided into 16 4 * 4;

Carrying out infra-frame prediction according to above-mentioned 9 kinds of predictor patterns respectively to each 4 * 4, serves as that optimum prediction subpattern is wherein selected on the basis with the SAD cost function, and the predicting the outcome as 4 * 4 of bases that predict the outcome of optimum prediction subpattern correspondence;

16 4 * 4 predict the outcome is combined into one 16 * 16, adopts Intra 4 * 4 intra prediction modes to carry out the result of infra-frame prediction as this MB; And each 4 * 4 cost function value additions under optimum prediction subpattern separately as the cost function value of this MB under Intra 4 * 4 intra prediction modes;

Then, adopt Intra 16 * 16 intra prediction modes that this MB is carried out infra-frame prediction again, carry out infra-frame prediction according to the predictor pattern of above-mentioned 4 kinds of permissions respectively, the cost function form class of its cost function and Intra 4 * 4 intra prediction modes seemingly, indivedual coefficient differences; Can also select for use (as a kind of situation of option) to be basis selection optimum prediction subpattern wherein according to H.264 stipulating in addition based on the cost function of Hadamard (Hadarmard) conversion coefficient, and the predicting the outcome of optimum prediction subpattern correspondence as the predicting the outcome of this MB, simultaneously the cost function value under the optimum prediction subpattern as the cost function value of this MB under Intra 16 * 16 intra prediction modes;

Cost function value the relation in an inequality of a more above-mentioned at last MB under Intra 4 * 4 intra prediction modes and Intra 16 * 16 intra prediction modes selected to use Intra 4 * 4 intra prediction modes or used Intra 16 * 16 intra prediction modes according to the comparative result of inequality.

In the selection course of above intra prediction mode, if obtained identical cost function value when MB or two kinds of predictor patterns of 4 * 4 employings, then choose the less predictor pattern of sequence number as optimum prediction subpattern (obtained identical cost function value when adopting Made0 with Made1, then can choose Made0) as the optimum prediction subpattern as MB.

Because the luminance component frame has 4 kinds of predictor patterns under Intra 16 * 16 intra prediction modes, 9 kinds of predictor patterns are arranged under Intra 4 * 4 intra prediction modes, and it is exactly to wish to select a kind of intra prediction mode to make that the code flow minimum and the signal to noise ratio of encoding out are the highest that frame mode is selected.The best way just is to use the Cost cost function to characterize the result of predictor pattern, and wherein the Cost cost function is based on the function that rate-distortion optimization (RDO=rate-distortion optimization) obtains in conjunction with SAD, specific as follows shown in:

Cost＝SAD+2Rλ(Qp)

In the following formula, SAD be absolute error and, promptly be reference picture and the difference of compilation as correspondence position.For the fine or not degree of more every kind of predictor pattern more accurately, get coefficients R=0 or R=1, for the minimum corresponding predictor pattern of trying to achieve SAD, get R=0; For other predictor pattern, get R=1.Wherein λ is the Lagrangian factor, λ=0.85*2 ^(Qp-12)/3, Qp is a quantization parameter in this formula, span is [0,51].

As an option of standard H.264, can also carry out Hadamard Hadarmard conversion to the infra-frame prediction difference, difference is transformed to frequency domain ask absolute error and SAD (because these differences will transform to frequency domain at last and encode), using the Hadarmard conversion here and not using discrete cosine transform (dct transform) mainly is to consider that the Hadarmard conversion realizes that principle is simple and relatively near dct transform.

According to above-mentioned, the corresponding brightness component frame is distinguished corresponding Intra 4 * 4 intra prediction modes and Intra16 * 16 intra prediction modes, defines its corresponding Cost cost function and is:

Cost _4×4＝SAD _4×4+2Rλ(Qp)

Cost _16×16＝SAD _16×16+2Rλ(Qp)

Correspondingly, to the colourity component frame, the Cost cost function of its corresponding Intra 8 * 8 intra prediction modes definition is:

Cost _8×8＝SAD _8×8+2Rλ(Qp)

For the luminance component frame, whether set up to decide and select which kind of intra prediction mode by differentiating following inequality:

Cost _16×16≥16Cost _4×4

If this inequality is set up, then select Intra 4 * 4 intra prediction modes to carry out infra-frame prediction, otherwise select Intra 16 * 16 intra prediction modes to carry out infra-frame prediction.

As fully visible, selecting for the intra prediction mode of luminance component frame is that the various predictor model prediction results that need compare under Inter 4 * 4 intra prediction modes and Intra 16 * 16 intra prediction modes each MB in the image finally determine selected intra prediction mode, therefore required amount of calculation is bigger, and the resource that consumes processor is also many.According to data statistics, H.264 in the compressed encoding standard, the selection link of intra prediction mode accounts for the 10%-12% of whole H.264 cataloged procedure institute's consumption calculations time, therefore, H.264 it is very necessary that the intra prediction mode selection course in the standard is optimized, to obtain high efficiency method for choosing frame inner forecast mode.

Summary of the invention

The technical problem to be solved in the present invention is to propose a kind of system of selection of intra prediction mode, to improve the efficiency of selection of intra prediction mode in the compression encoding process H.264, the multimedia communication performance is better improved.

For addressing the above problem, the present invention proposes a kind of system of selection of intra prediction mode, be used for the intra prediction mode of encoding video pictures process is selected, comprise step:

(1) chooses reference image block for the image block that will carry out infra-frame prediction in the current image frame;

(2) controlling this image block chooses corresponding optimum prediction subpattern and carries out infra-frame prediction in the formed predictor set of modes of the optimum prediction subpattern of each reference image block.

Described reference image block comprises: this will carry out the adjacent encoded image piece of the image block of infra-frame prediction; Reach the corresponding correspondence position image block in the position of image block in current image frame that in the previous image frame, will carry out infra-frame prediction with this; And the adjacent image piece of this correspondence position image block.

Wherein said adjacent encoded image piece is an adjacent top image block, adjacent left image block and adjacent upper left-hand image piece; The adjacent image piece of frame correspondence position image block is adjacent upper left-hand image piece before described, adjacent top image block, adjacent upper right side image block, adjacent left image block, adjacent right-hand image block, adjacent lower left image block, adjacent lower images piece and adjacent lower right image block.

The process of choosing the optimum prediction subpattern for this image block that will carry out infra-frame prediction in the described step (2) comprises step:

(21) make this image block that will carry out infra-frame prediction under the optimum prediction subpattern of each reference image block, carry out infra-frame prediction respectively, and calculate the Cost functional value that under every kind of optimum prediction subpattern, carries out infra-frame prediction respectively;

(22) the predictor pattern of choosing minimum function value correspondence will be carried out the optimum prediction subpattern of the image block of infra-frame prediction as this.

Described step (1) also comprises step before:

(A1) judge that whether current image frame is first picture frame in the sequence of video images, if go to step (A2); Otherwise go to step (1);

(A2) control each image block in this first picture frame, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose the optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes and carry out infra-frame prediction; Processing for subsequent image frames goes to step (1).

Described step (2) also comprises step afterwards: each image block in nT the picture frame of control of video image sequence, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose the optimum prediction subpattern in 4 kinds of predictor patterns under Intra16 * 16 and Intra 8 * 8 intra prediction modes and carry out infra-frame prediction; Described n is a natural number, and described T is determined by following formula:

T＝T ₀·λ(QP)

Wherein, λ (QP)=0.25 * 2 ^-(QP-12)/6, QP is a quantization parameter, T ₀=200.

Correspondingly, the invention allows for a kind of system of selection of intra prediction mode, be used for the intra prediction mode of encoding video pictures process is selected, comprising:

The fixed sequence program reference image block choose step:

(A) choose reference image block for the infra-frame prediction image block;

(B) calculate correlation between infra-frame prediction image block and the reference image block that each is chosen respectively;

(C) according to the size of infra-frame prediction image block, corresponding in all reference image blocks of choosing selected N fixed sequence program reference image block that correlation is high, described N is a natural number;

Image block infra-frame prediction step:

(D) follow-up carrying out in the infra-frame prediction process, according to the size of the infra-frame prediction image block in the current image frame, directly in the formed predictor set of modes of optimum prediction subpattern of corresponding selected fixed sequence program reference image block, choose corresponding optimum prediction subpattern and carry out infra-frame prediction.

Described step (A) and (B) between also comprise step: calculate respectively in infra-frame prediction image block and each reference image block 9 kinds of predictor patterns under Intra 16 * 16 intra prediction modes, or the optimum prediction subpattern in 4 kinds of predictor patterns under Intra8 * 8 or Intra 4 * 4 intra prediction modes;

The correlation of described step (B) is calculated according to following formula:

$\mathrm{Blkcorr}(f,p)=\frac{1}{\mathrm{MRC}}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{R-1}{\mathrm{\Σ}}}\underset{k=0}{\overset{C-1}{\mathrm{\Σ}}}p(i,j,k)$

Wherein:

(f is p) for identifying the parameter of reference image block;

(f p) is infra-frame prediction image block and with (f, p) relevance values between Biao Shi the reference image block to Blkcorr; Represent current image frame during described f=n, represent the previous image frame during f=n-1; Described p=0,1,2,3,4,5,6,7,8, the position of expression reference image block in picture frame;

M is the picture frame number that comprises in the sequence of video images, and R is the image block number of vertical direction in the picture frame, and C is the image block number of horizontal direction in the picture frame;

I is the frame number of infra-frame prediction image block place picture frame, 0≤i≤M-1; J is the row of infra-frame prediction image block in picture frame number, 0≤j≤R-1; K is the row of infra-frame prediction image block in picture frame number, 0≤k≤C-1;

When the infra-frame prediction image block and with (f, when p) the optimum prediction subpattern of Biao Shi reference image block is identical, p (i, j, k)=1;

When the infra-frame prediction image block and with (f, p) the optimum prediction subpattern of Biao Shi reference image block not simultaneously, p (i, j, k)=0.

Wherein in the step (C) when described image block is in the luminance component frame 4 * 4, described N=5; The high fixed sequence program reference image block of described 5 correlations is respectively the adjacent top image block and the adjacent left image block of infra-frame prediction image block, and the correspondence position image block in the previous image frame, and the adjacent right-hand image block and the adjacent lower images piece of this correspondence position image block; In the step (C) when described image block be in the luminance component frame 16 * 16 or during for 8 * 8 in the chromatic component frame, described N=3; The high fixed sequence program reference image block of described 3 correlations is respectively the adjacent top image block and the adjacent left image block of infra-frame prediction image block, and the correspondence position image block in the previous image frame.

The process of choosing the optimum prediction subpattern for the infra-frame prediction image block in the wherein said step (D) comprises step:

(D1) make the infra-frame prediction image block under the optimum prediction subpattern of corresponding selected fixed sequence program reference image block, carry out infra-frame prediction respectively, and calculate the Cost functional value that under every kind of optimum prediction subpattern, carries out infra-frame prediction respectively;

(D2) choose of the optimum prediction subpattern of the predictor pattern of minimum function value correspondence as the infra-frame prediction image block.

Described step (D) also comprises step before:

(P1) judge that whether current image frame is first picture frame in the sequence of video images, if go to step (P2); Otherwise go to step (D);

(P2) control each image block in this first picture frame, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose the optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes and carry out infra-frame prediction; Processing for subsequent image frames goes to step (D).

Described step (D) also comprises step afterwards: each image block in nT the picture frame of control of video image sequence, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose the optimum prediction subpattern in 4 kinds of predictor patterns under Intra16 * 16 and Intra 8 * 8 intra prediction modes and carry out infra-frame prediction; Described n is a natural number, and described T is determined by following formula:

T＝T ₀·λ(QP)

Wherein, λ (QP)=0.25 * 2 ^-(QP-12)/6, QP is a quantization parameter, T ₀=200.

The system of selection of intra prediction mode of the present invention is by choosing corresponding reference image block for the image block that will carry out infra-frame prediction, make this image block that will carry out infra-frame prediction in the predictor set of modes that the optimum prediction subpattern of the reference image block of choosing is formed, choose the optimum prediction subpattern of self, thereby can reduce the selected predictor pattern of the image block quantity that this will carry out infra-frame prediction, as for Intra 4 * 4, each 4 * 4 selectable predictor patterns will be less than 9 kinds, for Intra 16 * 16 or Intra8 * 8 intra prediction modes, each 16 * 16 or 8 * 8 selectable predictor patterns will be less than 4 kinds; Therefore improved the efficiency of selection of the intra prediction mode in the compression encoding process H.264 effectively, made the multimedia communication performance obtain better improvement.

Description of drawings

Fig. 1 is existing H.264 compression coding technology carries out the sequential compression coding to each macro block MB a schematic diagram;

Fig. 2 is that existing H.264 standard is carried out the prediction direction schematic diagram that infra-frame prediction limited;

Fig. 3 is the forecasting process schematic diagram that existing H.264 standard is carried out on each prediction direction shown in Figure 2 respectively;

Fig. 4 is the main realization principle flow chart of the system of selection of intra prediction mode of the present invention;

Fig. 5 is the schematic diagram of the selected reference image block of current image block in the system of selection of intra prediction mode of the present invention;

Fig. 6 determines the process chart of the fixed sequence program reference image block that correlation is stronger by increasing correlation calculations in the system of selection of intra prediction mode of the present invention;

Fig. 7 is in the system of selection of intra prediction mode of the present invention, for further improving the process chart that the infra-frame prediction accuracy is proposed.

Embodiment

In the process of the interior prediction processing of conducting frame, for 4 * 4 intra prediction modes of the Intra in the luminance component frame, each 4 * 4 will [be respectively Mode0 (vertically) prediction shown in Figure 39 kinds of predictor patterns, Mode1 (level) prediction, Mode2 (DC) prediction, Mode3 (left side tiltedly down) prediction, Mode4 (right side tiltedly down) prediction, Mode5 (vertically taking over) prediction, Mode6 (level on the lower side) prediction, Mode7 (vertically taking back) prediction and Mode8 (level is on the upper side) predict] in select self optimum prediction subpattern, for Intra 8 * 8 intra prediction modes in 16 * 16 intra prediction modes of the Intra in the luminance component frame and the chromatic component frame, each 16 * 16 or 8 * 8 will [for Intra 16 * 16 intra prediction modes, 4 kinds of predictor patterns be Mode0 (vertically) prediction 4 kinds of predictor patterns, Mode1 (level) prediction, Mode2 (DC) prediction and Mode3 (plane) prediction; For Intra 8 * 8 intra prediction modes, 4 kinds of predictor patterns are that Mode0 (DC) prediction, Mode1 (level) prediction, Mode2 (vertically) prediction and Mode3 (plane) predict] in select self optimum prediction subpattern, therefore the predictor pattern quantity of required comparison is bigger, and such its amount of calculation of predictor model selection mode obviously is sizable.

The system of selection of intra prediction mode of the present invention is at the defective of the above-mentioned existence of prior art, analysis draws in the process of prediction processing in the conducting frame, each image block (comprises 16 * 16 and 4 * 4 in the luminance component frame, and in the chromatic component frame 8 * 8) optimum prediction subpattern is relevant with the position at self place, also with self some image block on every side, and the image block on the preceding frame correspondence position, and the optimum prediction subpattern of some image blocks around the image block on the correspondence position is relevant, i.e. the optimum prediction subpattern of each image block can come unique decision by the optimum prediction subpattern of these associated picture pieces.And according to above-mentioned this analysis result, proposition utilizes the correlation of sequence of video images on time and space, according to image block around the current image block, and the image block of preceding frame correspondence position and on every side the attribute of piece carry out the Preliminary screening of predictor pattern, thereby exclude the very low predictor pattern of those possibilities, promptly do not need to differentiate the predictor pattern of these low possibilities, and only need differentiate the predictor pattern of high likelihood; So just can reduce the predictor model number that each image block need be differentiated, exclude the very low predictor pattern of some obvious possibilities, only possibility is differentiated than higher predictor pattern then,, reached purpose of the present invention to reduce the amount of calculation of predictor model selection.

In compressed encoding standard H.264, to carry out infra-frame prediction to each macro block MB during frame in the coded frame, in predictive frame, also to carry out infra-frame prediction and inter prediction simultaneously to each macro block MB, select to finish predictive mode.Especially should do Intra 4 * 4 infra-frame predictions for each image block in the luminance component frame, also to do Intra 16 * 16 infra-frame predictions, therefore be the present invention improve the intra prediction mode efficiency of selection at emphasis, to reduce the H.264 computational complexity of encoder, satisfy the real-time requirement of multimedia communication.

Be explained in detail below in conjunction with the embodiment of each accompanying drawing the system of selection of intra prediction mode of the present invention.See also Fig. 4, this figure is the main realization principle flow chart of the system of selection of intra prediction mode of the present invention, and its main implementation procedure is:

Step S10 is for the image block that will carry out infra-frame prediction in the current image frame is chosen reference image block; The reference image block of wherein choosing can comprise:

This will carry out the adjacent encoded image piece of the image block of infra-frame prediction, coding rule according to video image blocks, described adjacent encoded image piece comprises the adjacent top image block of the image block that this will carry out infra-frame prediction, adjacent left image block and adjacent upper left-hand image piece; And

In the previous image frame, to carry out the corresponding correspondence position image block in the position of image block in current image frame of infra-frame prediction with this; And

The adjacent image piece of this correspondence position image block, wherein the adjacent image piece of correspondence position image block comprises its adjacent upper left-hand image piece, adjacent top image block, adjacent upper right side image block, adjacent left image block, adjacent right-hand image block, adjacent lower left image block, adjacent lower images piece and 8 image blocks of adjacent lower right image block.

Step S20, control this image block that will carry out infra-frame prediction and choose corresponding optimum prediction subpattern carry out infra-frame prediction in the formed predictor set of modes of the optimum prediction subpattern of each reference image block of choosing, the process that this image block that will carry out infra-frame prediction is chosen the optimum prediction subpattern is as follows:

Make this image block that will carry out infra-frame prediction under the optimum prediction subpattern of each reference image block, carry out infra-frame prediction respectively; And

This image block that will carry out infra-frame prediction carries out the functional value of infra-frame prediction under the optimum prediction subpattern of each reference image block based on the Cost cost function calculation respectively; And

Choose the predictor pattern of minimum function value correspondence and will carry out the optimum prediction subpattern of the image block of infra-frame prediction as this.

The system of selection of intra prediction mode of the present invention realizes that the key of technology is under Intra16 * 16 and Intra 4 * 4 two kind of intra prediction mode and chromatic component frame Intra 8 * 8 intra prediction modes of luminance component frame, reducing each image block as much as possible (can be macro block MB in the luminance component frame or 4 * 4, and in the chromatic component frame 8 * 8) need the quantity of the predictor pattern of computational discrimination, but the accuracy that can not significantly reduce infra-frame prediction (generally speaking, efficient processing mode is with respect to original method, its overall performance always has some reductions, but should guarantee that this reduction amount should be as far as possible little, not remarkable in other words).Its main thought of the system of selection of intra prediction mode of the present invention is: utilize previous image frame and current image frame correspondence position image block and the infra-frame prediction subpattern of image block on every side thereof to predict the optimum prediction subpattern that will carry out the image block of infra-frame prediction in the current image frame, thereby reduce the predictor pattern quantity that current image block needs computational discrimination.

See also Fig. 5, this figure is the schematic diagram of the selected reference image block of current image block in the system of selection of intra prediction mode of the present invention; Suppose in Fig. 5, be to the M in the current image frame (N frame) ₀ ⁿImage block (can be in the luminance component frame 4 * 4 or 16 * 16 and determine, or in the chromatic component frame 8 * 8) carries out infra-frame prediction, then the image block M in the N picture frame ₀ ⁿThe selection course of intra prediction mode be: by the image block M in the N-1 picture frame ₀ ^N-1, M ₁ ^N-1..., M ₈ ^N-1Image block M in 9 image blocks and the N picture frame ₁ ⁿ, M ₂ ⁿAnd M ₃ ⁿInfra-frame prediction is carried out in its optimum prediction subpattern of computational discrimination in the formed predictor set of modes of the optimum prediction subpattern of image block.Like this, for M ₀ ⁿThe Preliminary screening mode of the predictor model selection of image block is to think current image block M ₀ ⁿThe optimum prediction subpattern of 12 reference image block correspondences are the higher candidate's predictor patterns of possibility, need relatively differentiate by calculating respectively, and for the predictor pattern in these reference image block scopes not, just being considered as is the lower predictor pattern of possibility, can directly give up the processing that will not calculate.Simultaneously because in 12 reference image blocks, the optimum prediction subpattern that has a plurality of reference image blocks is identical, therefore corresponding optimum prediction subpattern number may lack than 9 kinds of the prior art (for for 4 * 4 intra prediction modes of the Intra in the luminance component frame) or 4 kinds of (for Intra 8 * 8 intra prediction modes in 16 * 16 intra prediction modes of the Intra in the luminance component frame or the chromatic component frame) predictor model numbers, so just reached and reduced the purpose of differentiating amount of calculation, promptly made the M that will carry out infra-frame prediction in the N picture frame ₀ ⁿImage block needs the infra-frame prediction subpattern number of computational discrimination will be far fewer than 9 kinds or 4 kinds of the prior art, thereby has improved the efficiency of selection of intra prediction mode greatly.

Wherein, the M in the N-1 frame among Fig. 5 ₀ ^N-1, M ₁ ^N-1, M ₂ ^N-1, M ₃ ^N-1, M ₄ ^N-1, M ₅ ^N-1, M ₆ ^N-1, M ₇ ^N-1, M ₈ ^N-1M in 9 image blocks and the N frame ₁ ⁿ, M ₂ ⁿ, M ₃ ^N3Individual image block is this image block M that will carry out infra-frame prediction ₀ ⁿReference image block; M wherein ₁ ⁿ, M ₂ ⁿAnd M ₃ ⁿImage block is image block M ₀ ⁿAdjacent encoded image piece, M ₀ ^N-1Image block is image block M ₀ ⁿThe correspondence position image block in the previous image frame, M ₁ ^N-1, M ₂ ^N-1, M ₃ ^N-1, M ₄ ^N-1, M ₅ ^N-1, M ₆ ^N-1, M ₇ ^N-1, M ₈ ^N-1Be correspondence position image block M ₀ ^N-1The adjacent image piece.According to the coding rule of compressed encoding H.264, these reference image blocks all will be prior to current image block M ₀ ⁿFinish encoding process, therefore all be can reference image block.

For further reducing the computational discrimination complexity of infra-frame prediction subpattern, the system of selection of the intra prediction mode of the present invention image block around current image block and its of also giving chapter and verse, and preceding frame correspondence position image block and on every side the strong and weak degree of correlation of image block sort and determine the fixed sequence program reference image block of some stronger reference image blocks of correlation as the infra-frame prediction image block, directly predict the optimum prediction subpattern of predicted picture piece in the present frame then according to the attribute of the stronger fixed sequence program reference image block of these correlations,, and further reduce the amount of calculation that the predictor model selection is handled so that but the predictor pattern of computational discrimination further screened.

See also Fig. 6, this figure determines the process chart of the fixed sequence program reference image block that correlation is stronger by increasing correlation calculations in the system of selection of intra prediction mode of the present invention; Its main implementation procedure is as follows:

Step S21 is respectively at the image sequence of different motion attribute, for the infra-frame prediction image block is chosen reference image block; Wherein the reference image block of choosing for the infra-frame prediction image block also can be the M in the N-1 frame shown in Fig. 5 ₀ ^N-1, M ₁ ^N-1, M ₂ ^N-1, M ₃ ^N-1, M ₄ ^N-1, M ₅ ^N-1, M ₆ ^N-1, M ₇ ^N-1, M ₈ ^N-1M in 9 image blocks and the N frame (being present frame) ₁ ⁿ, M ₂ ⁿ, M ₃ ⁿ3 image blocks.

Step S22, calculate the infra-frame prediction image block respectively, and in 9 kinds of predictor patterns of each reference image block of choosing under Intra 16 * 16 intra prediction modes, or the optimum prediction subpattern in 4 kinds of predictor patterns under Intra 8 * 8 or Intra 4 * 4 intra prediction modes;

Step S23 calculates the correlation between infra-frame prediction image block and the reference image block that each is chosen respectively; Wherein said correlation can be calculated according to following formula:

$\mathrm{Blkcorr}(f,p)=\frac{1}{\mathrm{MRC}}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{R-1}{\mathrm{\Σ}}}\underset{k=0}{\overset{C-1}{\mathrm{\Σ}}}p(i,j,k)$

In the following formula: (f is p) for identifying the parameter of reference image block;

(f p) will carry out the image block of infra-frame prediction for this and with (f, p) relevance values between Biao Shi the reference image block to Blkcorr; Wherein represent current image frame during f=n, represent the previous image frame during f=n-1; Described p=0,1,2,3,4,5,6,7,8, the position of expression reference image block in picture frame;

M is the number of the picture frame that comprises in the sequence of video images, and R is the image block number of vertical direction in the picture frame, and C is the image block number of horizontal direction in the picture frame;

I will carry out the frame number of the image block place picture frame of infra-frame prediction, 0≤i≤M-1 for this; J will carry out the row of image block in picture frame number of infra-frame prediction, 0≤j≤R-1 for this; K will carry out the row of image block in picture frame number of infra-frame prediction, 0≤k≤C-1 for this;

Wherein when this to carry out infra-frame prediction image block and with (f, when p) the optimum prediction subpattern of Biao Shi reference image block is identical, then p (i, j, k)=1; When this to carry out infra-frame prediction image block and with (f, p) the optimum prediction subpattern of Biao Shi reference image block not simultaneously, then p (i, j, k)=0.

Step S24, according to the size of infra-frame prediction image block, corresponding individual (N is a natural number) the fixed sequence program reference image block that correlation is high of N of in all reference image blocks of choosing, selecting;

As, when the selected image block of infra-frame prediction is in the luminance component frame 4 * 4, can select the fixed sequence program reference image block of the higher image block of 5 (being N=5) correlations as the infra-frame prediction image block; The higher image block of wherein choosing of 5 correlations can be respectively the adjacent top image block and the adjacent left image block of the image block that will carry out infra-frame prediction, and the correspondence position image block in the previous image frame, and the adjacent right-hand image block and the adjacent lower images piece of this correspondence position image block.

And for example, when the selected image block of infra-frame prediction is in the luminance component frame 16 * 16 or during for 8 * 8 in the chromatic component frame, can select the fixed sequence program reference image block of the higher image block of 3 (being N=3) correlations as the infra-frame prediction image block; Wherein 3 higher image blocks of correlation can be respectively the adjacent top image block and the adjacent left image block of the image block that will carry out infra-frame prediction, and the correspondence position image block in the previous image frame.

Basis based on above-mentioned result, for follow-up intra-prediction process, can be according to the size of infra-frame prediction image block, and according to the determined fixed sequence program reference image block of step S24, directly choose the higher fixed sequence program reference image block of a corresponding N correlation (as for 4 * 4 in the luminance component frame, can directly select to carry out the adjacent top image block and the adjacent left image block of the image block of infra-frame prediction, and the correspondence position image block in the previous image frame, and 5 image blocks such as the adjacent right-hand image block of this correspondence position image block and adjacent lower images piece will carry out the reference image block of the image block of infra-frame prediction as this; And for 16 * 16 in the luminance component frame or be in the chromatic component frame 8 * 8, can directly choose the adjacent top image block and the adjacent left image block of the image block that will carry out infra-frame prediction, and 3 image blocks such as correspondence position image block in the previous image frame will carry out the reference image block of the image block of infra-frame prediction as this, so just can reduce the amount of calculation of correlation to a great extent), in the formed predictor set of modes of optimum prediction subpattern of the higher fixed sequence program reference image block of the N that a chooses correlation, choose corresponding optimum prediction subpattern then and carry out infra-frame prediction.

Because consider current image block M from the correlation angle of sequence of video images ₀ ⁿWith the correlation of its 12 reference image blocks be different, some correlation is than higher, such as reference image block M ₀ ^N-1The other correlation is lower, such as reference image block M ₆ ^N-1Therefore, just can be by the correlation between each reference image block and the current image block that will carry out infra-frame prediction is sorted, with the number of further screening infra-frame prediction subpattern.

Based on above-mentioned principle, the present invention proposes 4 * 4 M that will carry out infra-frame prediction in the current image frame ₀ ⁿThe optimum prediction subpattern and the previous image frame in 4 * 4 M of correspondence position ₀ ^N-1, and piece M ₀ ^N-1Adjacent block M ₁ ^N-1, M ₂ ^N-1, M ₃ ^N-1, M ₄ ^N-1, M ₅ ^N-1, M ₆ ^N-1, M ₇ ^N-1And M ₈ ^N-1, and the adjacent block M in the current image frame ₁ ⁿ, M ₂ ⁿAnd M ₃ ⁿStatistical comparison is carried out in the optimum prediction subpattern of totally 12 reference image blocks, and provides each reference image block and current image block M ₀ ⁿCorrelation, the correlation here is meant " the optimum prediction subpattern of certain reference image block and current image block M ₀ ⁿThe optimum prediction subpattern identical " probability of this incident.Wherein, the concrete computational methods of correlation are as follows:

$\mathrm{Blkcorr}(f,p)=\frac{1}{\mathrm{MRC}}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{R-1}{\mathrm{\Σ}}}\underset{k=0}{\overset{C-1}{\mathrm{\Σ}}}p(i,j,k),$

Wherein

Wherein, (f p) has characterized current image block M to Blkcorr ₀ ⁿWith reference image block M _p ^fBetween correlation; F=n (expression current image frame) or n-1 (expression previous image frame); P=0,1,2,3,4,5,6,7,8, the location label of presentation video piece in picture frame; M is an all images frame number in the sequence of video images, and R represents the image block number of vertical direction in the picture frame, and C represents the image block number of horizontal direction in the picture frame.In the following formula summation process, (i, j k), are sign current image block M for summing target ₀ ⁿJ in the i frame is capable, the image block of k row.Before the correlation of certain sequence of video images of aforementioned calculation, the selection of be respectively each image block in this sequence of video images being carried out the optimum prediction subpattern according to 9 kinds of predictor patterns of the prior art or 4 kinds of predictor patterns is handled.

In order to make the extensive sequence of video images of the above-mentioned this correlation scope of application, the present invention calculates the representative image sequence of various typical types by repeatedly testing, and has provided the correlation data of these three typical image sequences, and is as shown in the table:

Each reference image block of table 1. and M ₀ ⁿThe comparison of correlation

Image sequence	M 0 n-1	M 1 n-1	M 2 n-1	M 3 n-1	M 4 n-1	M 5 n-1	M 6 n-1	M 7 n-1	M 8 n-1	M 1 n	M 2 n	M 3 n
													foreman Claire news	0.549 0.749 0.801	0.339 0.546 0.465	0.349 0.557 0.477	0.356 0.497 0.475	0.384 0.520 0.485	0.297 0.406 0.359	0.277 0.428 0.367	0.270 0.417 0.369	0.327 0.438 0.376	0.421 0.624 0.496	0.464 0.584 0.518	0.349 0.450 0.373

As can be seen from Table 1, current image block M ₀ ⁿOptimum prediction subpattern and reference image block M ₀ ^N-1Correlation the highest, secondly be and reference image block M ₁ ⁿ, M ₂ ⁿCorrelation higher, be once more and reference image block M ₁ ^N-1, M ₂ ^N-1, M ₃ ^N-1, and M ₄ ^N-1Correlation take second place, with its correlation lower be reference image block M ₅ ^N-1, M ₆ ^N-1, M ₇ ^N-1, M ₈ ^N-1And M ₃ ⁿTherefore, in order to make full use of the correlation of sequence of video images on room and time, and reducing the quantity selected of predictor pattern as far as possible, the present invention chooses here and uses the optimum prediction subpattern of the higher fixed sequence program reference image block of following 5 correlations to predict current image block M ₀ ⁿOptimum prediction subpattern (please consulting the shadow image piece among Fig. 5 simultaneously):

1) 4 * 4 image block M of correspondence position in the previous image frame (i.e. N-1 frame) ₀ ^N-1, and this 4 * 4 image block M ₀ ^N-1Adjacent right-hand image block M ₄ ^N-1With adjacent lower images piece M ₂ ^N-1

2) the current image block M in the current image frame (i.e. N frame) ₀ ⁿAdjacent top image block M ₁ ⁿWith adjacent left image block M ₂ ⁿ

Simultaneously, the optimum prediction subpattern of the reference image block that 5 selected here correlations are higher has identical possibility to a great extent, so current image block M ₀ ⁿThe number of the infra-frame prediction selected subpattern will be less than 5 kinds.After table 2 has provided and has used the system of selection of intra prediction mode of the present invention, the accuracy parameter of infra-frame prediction and the average required predictor model selection number of doing are as follows:

The comparison (QP=28) of table 2. prediction accuracy and average predictive mode number

Image sequence (QCIF)	Prediction accuracy			Image sequence (QCIF)	Consensus forecast subpattern number
								Brightness 4 * 4	Brightness 16 * 16	Colourity 8 * 8	Brightness 4 * 4	Brightness 16 * 16	Colourity 8 * 8
	foreman claire news	85.7％ 94.3％ 93.5％	93.2％ 95.2％ 97.4％		88.8％ 94.4％ 95.2％	foreman claire news	2.79 2.27 2.54							2.59 2.47 2.44	2.58 2.52 2.52

Above-mentioned processing procedure and the effect that adopts the system of selection of intra prediction mode of the present invention at 4 * 4 intra prediction modes of the Intra in the luminance component frame of having narrated, compare with Intra 4 * 4 intra prediction modes in the luminance component frame, selectable predictor number of modes after the system of selection of Intra 16 * 16 intra prediction modes in the luminance component frame and the Intra8 in the chromatic component frame * 8 intra prediction modes employing intra prediction mode of the present invention will reduce half than 4 kinds of prior art, so, only need to current image block M for luminance component frame Intra 16 * 16 intra prediction modes and chromatic component frame Intra 8 * 8 intra prediction modes ₀ ⁿChoose image block M ₀ ^N-1, M ₁ ⁿAnd M ₂ ⁿAs its fixed sequence program reference image block, promptly with image block M ₀ ^N-1, M ₁ ⁿAnd M ₂ ⁿThe optimum prediction subpattern as candidate's predictor pattern, in this predictor model domain, carry out computational discrimination and choose current image block M ₀ ⁿThe optimum prediction subpattern of self, the accuracy of its infra-frame prediction and average selectable predictor number of modes thereof see also table 2 record.

Utilize the result of the respective fixation sequence reference image block that aforementioned calculation obtains then, follow-up when image block being carried out infra-frame prediction at every turn, just can be according to the size of infra-frame prediction image block, the respective fixation sequence reference image block that directly uses above-mentioned repeatedly experiment to obtain, and in the formed predictor set of modes of optimum prediction subpattern of the fixed sequence program reference image block of determining, choose corresponding optimum prediction subpattern and carry out infra-frame prediction.

By last table 2 as can be seen, the system of selection of intra prediction mode of the present invention can guarantee that the infra-frame prediction accuracy reaches on the quite high basis, make each image block on average need the predictor model number of computational discrimination to reduce many in terms of existing technologies, particularly for 4 * 4 intra prediction modes of the Intra in the luminance component frame, need the predictor number of modes of computational discrimination to be reduced to about 2.5 kinds by original 9 kinds, the reduction of its computation complexity is very obvious.

Though after the system of selection of enforcement intra prediction mode of the present invention, its infra-frame prediction accuracy is quite high, compares with prior art, still has certain error; And therefore this error can will cause the accumulation of this error to amplify along with the picture frame backward progressively diffusion propagation of cataloged procedure from the front.So for diffusion and the accumulation of eliminating this error, when the present invention proposes to utilize the inventive method to carry out the intra prediction mode selection, regulation can not be used method for choosing frame inner forecast mode of the present invention to some particular image frame, but still use the intra prediction mode selection mode of prior art, to improve the accuracy of infra-frame prediction, prevent the diffusion and the propagation of error, wherein these particular image frames and processing mode thereof are as follows:

At first, can not use the selection algorithm of intra prediction mode of the present invention to each image block in first picture frame of sequence of video images, in must 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or select the processing of its optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes, so that the infra-frame prediction of subsequent image frames is more accurate.

Once more, in the whole video image sequence, need select a picture frame F every certain frame number T _Ecc, to the picture frame F of this selection _EccIn each image block, in must 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or select the processing of its optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes, adopt the infra-frame prediction error after the system of selection of intra prediction mode of the present invention with thorough elimination prior figures picture frame.Wherein choosing of T value can be determined according to following formula:

T＝T ₀·λ(QP)

In the following formula, λ (QP)=0.25 * 2 ^-(QP-12)/6, wherein QP is a quantization parameter, T ₀For QP equals the value of 0 o'clock T, wherein T ₀=200.

See also Fig. 7, this figure is in the system of selection of intra prediction mode of the present invention, for further improving the process chart that the infra-frame prediction accuracy is proposed; Its main processing procedure is as follows:

Step S31 judges that whether current image frame is first picture frame in the sequence of video images, if, execution in step S32; Otherwise execution in step S33;

Step S32, control each image block in this first picture frame, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose its optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes and carry out infra-frame prediction; Processing for subsequent image frames then goes to execution in step S33;

Step S33 is for infra-frame prediction image block in the current image frame is chosen reference image block;

Step S34, the interior predicted picture piece of control frame is chosen corresponding optimum prediction subpattern and is carried out infra-frame prediction in the formed predictor set of modes of the optimum prediction subpattern of each reference image block of choosing;

Step S35, each image block in the control of video image sequence in nT picture frame, in 9 kinds of predictor patterns under Intra4 * 4 intra prediction modes, or infra-frame prediction is carried out in the optimum prediction subpattern of choosing self in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes.Wherein determining as mentioned above of T repeats no more n=1,2,3,4...... here.

In sum, the multimedia telecom equipment of system of selection manufacturing that adopts intra prediction mode of the present invention is for the various multimedia communication products that adopt international standard H.264, can reduce the amount of calculation of video coding, thereby can be implemented under certain cost, improve the handling property of product, such as the frame per second that improves video image, perhaps the processing resource of processor is used for the processing that other calculates link, thereby improves the competitiveness of product in market; Perhaps, can improve the competitiveness of product in market too keeping significantly reducing input cost under the handling property condition suitable with like product.

Table 3. technical solution of the present invention and prior art scheme are in the contrast of aspect of performance

Image sequence	QP	PSNR(dB)		Bit rate (kbps)		Coding rate (f/s)		Coding rate improves (%)
									Existing	The present invention	Existing	The present invention	Existing	The present invention
		foreman foreman foreman claire claire claire news news news	18 28 38 18 28 38 18 28 38	44.199 36.719 29.894 47.347 40.628 33.729 45.173 37.527 29.837	44.176 36.691 29.852 47.326 40.593 33.696 45.171 37.519 29.835	1870.0 765.60 272.45 891.7 395.51 153.21 1714.2 829.57 334.20	1908.1 790.53 277.07 906.1 400.93 155.04 1734.7 838.83 335.58								34.0 39.9 45.5 39.7 40.7 45.6 35.0 39.4 43.7	39.3 46.9 49.9 45.8 50.1 51.2 40.5 46.3 49.0	15.6 17.5 9.6 15.4 23.1 12.3 15.7 17.5 12.1

According to technical solution of the present invention that last table 3 provided and prior art scheme in the reduced parameter aspect the handling property, can draw: for different typical test video image sequences, such as Foreman (strenuous exercise) image sequence, News (middle motion) image sequence and Claire (harmonic motion) image sequence are under different quantization parameter QP, and its handling property comprises:

PSNR (its PSNR value is high more, and picture quality is good more);

Coding produces the bit rate (its bit rate is low more, and compression efficiency is high more) of code stream; With

Coding rate (coding rate is high more, the level and smooth more smoothness of image motion sense).

By table 3 as seen, under the essentially identical situation of PSNR, the result that the result that technical solution of the present invention is handled handles with respect to the prior art scheme, its bit rate can slightly increase, but it is very little to increase numerical value, and (reason that bit rate increases is that the optimum prediction subpattern that the parts of images piece is selected departs from its real optimum prediction subpattern after adopting the system of selection of intra prediction mode of the present invention to be lower than 2% substantially, cause prediction residual to rise, cause that compression efficiency slightly descends); Yet the result that the result that technical solution of the present invention is handled handles with respect to the prior art scheme, coding rate has significantly and improves, coding rate has improved 10% to 20% with respect to the result of prior art scheme, improved the intraframe predictive coding speed in the compression encoding process H.264 greatly, the lifting of this performance is quite significant thus.If employing technical solution of the present invention, keep its code rate certain, such as remaining on 25fps, the disposal ability of the digital signal processor of so much media communication (DSP) can be saved about 15%, this part saved disposal ability can be used for handling the master control or the audio coding decoding of multimedia telecom equipment fully, thereby just can save an other chip, directly to cause the reduction of product cost.

The above only is a preferred implementation of the present invention; should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (15)

1, a kind of system of selection of intra prediction mode is used for the intra prediction mode of encoding video pictures process is selected, and it is characterized in that, comprises step:

(1) chooses reference image block for the image block that will carry out infra-frame prediction in the current image frame;

(2) controlling this image block chooses corresponding optimum prediction subpattern and carries out infra-frame prediction in the formed predictor set of modes of the optimum prediction subpattern of each reference image block.

2, the system of selection of intra prediction mode according to claim 1 is characterized in that, described reference image block comprises:

This will carry out the adjacent encoded image piece of the image block of infra-frame prediction; And

In the previous image frame, to carry out the corresponding correspondence position image block in the position of image block in current image frame of infra-frame prediction with this; And

The adjacent image piece of this correspondence position image block.

3, the system of selection of intra prediction mode according to claim 2 is characterized in that,

Described adjacent encoded image piece is an adjacent top image block, adjacent left image block and adjacent upper left-hand image piece;

The adjacent image piece of frame correspondence position image block is adjacent upper left-hand image piece before described, adjacent top image block, adjacent upper right side image block, adjacent left image block, adjacent right-hand image block, adjacent lower left image block, adjacent lower images piece and adjacent lower right image block.

4, according to the system of selection of claim 1,2 or 3 described intra prediction modes, it is characterized in that choosing the optimum prediction subpattern for this image block that will carry out infra-frame prediction in the described step (2) is to choose according to the Cost functional value of image block under every kind of optimum prediction subpattern that will carry out infra-frame prediction.

5, the system of selection of intra prediction mode according to claim 4 is characterized in that, the described process of choosing specifically comprises step:

(21) make this image block that will carry out infra-frame prediction under the optimum prediction subpattern of each reference image block, carry out infra-frame prediction respectively, and calculate the Cost functional value that under every kind of optimum prediction subpattern, carries out infra-frame prediction respectively;

(22) the predictor pattern of choosing minimum function value correspondence will be carried out the optimum prediction subpattern of the image block of infra-frame prediction as this.

According to the system of selection of claim 1,2 or 3 described intra prediction modes, it is characterized in that 6, described step (1) also comprises step before:

(A1) judge that whether current image frame is first picture frame in the sequence of video images, if go to step (A2); Otherwise go to step (1);

(A2) control each image block in this first picture frame, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose the optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes and carry out infra-frame prediction; Processing for subsequent image frames goes to step (1).

7, the system of selection of intra prediction mode according to claim 6, it is characterized in that, described step (2) also comprises step afterwards: each image block in nT the picture frame of control of video image sequence, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose the optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra8 * 8 intra prediction modes and carry out infra-frame prediction; Described n is a natural number, and described T is determined by following formula:

T＝T ₀·λ(QP)

Wherein, λ (QP)=0.25 * 2 ^-(QP-12)/6, QP is a quantization parameter, T ₀=200.

8, a kind of system of selection of intra prediction mode is used for the intra prediction mode of encoding video pictures process is selected, and it is characterized in that, comprising:

The fixed sequence program reference image block choose step:

(A) choose reference image block for the infra-frame prediction image block;

(B) calculate correlation between infra-frame prediction image block and the reference image block that each is chosen respectively;

(C) according to the size of infra-frame prediction image block, corresponding in all reference image blocks of choosing selected N fixed sequence program reference image block that correlation is high, described N is a natural number;

Image block infra-frame prediction step:

(D) follow-up carrying out in the infra-frame prediction process, according to the size of the infra-frame prediction image block in the current image frame, directly in the formed predictor set of modes of optimum prediction subpattern of corresponding selected fixed sequence program reference image block, choose corresponding optimum prediction subpattern and carry out infra-frame prediction.

9, the system of selection of intra prediction mode according to claim 8 is characterized in that, described reference image block comprises:

The adjacent encoded image piece of infra-frame prediction image block; And

The correspondence position image block corresponding in the previous image frame with the position of infra-frame prediction image block in current image frame; And

The adjacent image piece of this correspondence position image block.

10, the system of selection of intra prediction mode according to claim 9 is characterized in that,

Described adjacent encoded image piece is an adjacent top image block, adjacent left image block and adjacent upper left-hand image piece;

The adjacent image piece of frame correspondence position image block is adjacent upper left-hand image piece before described, adjacent top image block, adjacent upper right side image block, adjacent left image block, adjacent right-hand image block, adjacent lower left image block, adjacent lower images piece and adjacent lower right image block.

11, according to Claim 8, the system of selection of 9 or 10 described intra prediction modes, it is characterized in that, described step (A) and (B) between also comprise step:

Calculate respectively in infra-frame prediction image block and each reference image block 9 kinds of predictor patterns under Intra 16 * 16 intra prediction modes, or the optimum prediction subpattern in 4 kinds of predictor patterns under Intra 8 * 8 or Intra 4 * 4 intra prediction modes;

The correlation of described step (B) is calculated according to following formula:

$\mathrm{Blkcorr}(f,p)=\frac{1}{\mathrm{MRC}}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{R-1}{\mathrm{\Σ}}}\underset{k=0}{\overset{C-1}{\mathrm{\Σ}}}p(i,j,k)$

Wherein:

(f is p) for identifying the parameter of reference image block;

(f p) is infra-frame prediction image block and with (f, p) relevance values between Biao Shi the reference image block to Blkcorr; Represent current image frame during described f=n, represent the previous image frame during f=n-1; Described p=0,1,2,3,4,5,6,7,8, the position of expression reference image block in picture frame;

M is the picture frame number that comprises in the sequence of video images, and R is the image block number of vertical direction in the picture frame, and C is the image block number of horizontal direction in the picture frame;

I is the frame number of infra-frame prediction image block place picture frame, 0≤i≤M-1; J is the row of infra-frame prediction image block in picture frame number, 0≤j≤R-1; K is the row of infra-frame prediction image block in picture frame number, 0≤k≤C-1;

When the infra-frame prediction image block and with (f, when p) the optimum prediction subpattern of Biao Shi reference image block is identical, p (i, j, k)=1;

When the infra-frame prediction image block and with (f, p) the optimum prediction subpattern of Biao Shi reference image block not simultaneously, p (i, j, k)=0.

12, the system of selection of intra prediction mode according to claim 11 is characterized in that,

In the step (C) when described image block is in the luminance component frame 4 * 4, described N=5; The high fixed sequence program reference image block of described 5 correlations is respectively the adjacent top image block and the adjacent left image block of infra-frame prediction image block, and the correspondence position image block in the previous image frame, and the adjacent right-hand image block and the adjacent lower images piece of this correspondence position image block;

In the step (C) when described image block be in the luminance component frame 16 * 16 or during for 8 * 8 in the chromatic component frame, described N=3; The high fixed sequence program reference image block of described 3 correlations is respectively the adjacent top image block and the adjacent left image block of infra-frame prediction image block, and the correspondence position image block in the previous image frame.

13, the system of selection of intra prediction mode according to claim 8 is characterized in that, the process of choosing the optimum prediction subpattern for the infra-frame prediction image block in the described step (D) comprises step:

(D1) make the infra-frame prediction image block under the optimum prediction subpattern of corresponding selected fixed sequence program reference image block, carry out infra-frame prediction respectively, and calculate the Cost functional value that under every kind of optimum prediction subpattern, carries out infra-frame prediction respectively;

(D2) choose of the optimum prediction subpattern of the predictor pattern of minimum function value correspondence as the infra-frame prediction image block.

14, the system of selection of intra prediction mode according to claim 8 is characterized in that, described step (D) also comprises step before:

(P1) judge that whether current image frame is first picture frame in the sequence of video images, if go to step (P2); Otherwise go to step (D);

(P2) control each image block in this first picture frame, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose the optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes and carry out infra-frame prediction; Processing for subsequent image frames goes to step (D).

15, the system of selection of intra prediction mode according to claim 14, it is characterized in that, described step (D) also comprises step afterwards: each image block in nT the picture frame of control of video image sequence, in 9 kinds of predictor patterns under Intra 4 * 4 intra prediction modes, or choose the optimum prediction subpattern in 4 kinds of predictor patterns under Intra 16 * 16 and Intra 8 * 8 intra prediction modes and carry out infra-frame prediction; Described n is a natural number, and described T is determined by following formula:

T＝T ₀·λ(QP)

Wherein, λ (QP)=0.25 * 2 ^-(QP-12)/6, QP is a quantization parameter, T ₀=200.

Images