CN102625101A - Method for rapidly selecting H.264 interframe predication mode based on random pixel method - Google Patents

Abstract

The invention provides a method for rapidly selecting an H.264 interframe predication mode based on a random pixel method. The method mainly comprises the steps of: 1, randomly selecting a plurality of pixel points of a current image, and solving a variance yield delta of brightness components of all the pixel points; 2, setting a large dimension threshold TH1 and a small size threshold TH2; and 3, comparing the relationships between delta and the thresholds TH1 and TH2, and determining an optimal interframe prediction mode. Under the conditions that the selected number of the pixel points is proper and the thresholds TH1 and TH2 are set properly, the method can be used for rapidly determining the optimal prediction mode for interframe prediction, so that the prediction time is saved and the coding efficiency is increased effectively.

Description

A kind of H.264 inter-frame forecast mode fast selecting method based on the random pixel method

Technical field

The invention belongs to the multimedia video coding field, be specifically related to a kind of H.264 fast selecting method of inter-frame forecast mode that is directed against.Select suitable and dimension threshold is provided with under the suitable situation in the number of pixel, the present invention can accelerate the inter prediction process, improves code rate.

Background technology

H.264 be the video encoding standard of joint video team (JVT/Joint Video Team) exploitation of MPEG (Motion Picture Experts Group) of VCEG (video coding expert group) and the ISO/IEC of ITU-T.It is equal to the tenth part of the MPEG-4 of ISO, in official approval in May, 2003.Compare with traditional video coding mode, H.264 following advantage is arranged: the code efficiency that (1) is higher; (2) high-quality video picture; (3) adaptive time-delay characteristics; (4) mistake restore funcitons.

In cataloged procedure H.264, estimation is an operand the best part in the whole coding module.Therefore no matter be to realize or in the process that software is realized, to estimation with backoff algorithm is optimized and improvement is very important at hardware.In motion estimation module, how can obtain high-precision motion vector fast and effectively is the key that realizes high efficient coding.In present block matching algorithm, full-search algorithm (FS) has the highest search precision, but its operand is big, the drawbacks limit of real-time difference the application of FS in reality.

H.264 adopt rate-distortion optimization (Rate Distortion Optimization; RDO) judge that selection divides block mode; Encoder is to all possible pattern difference of each macro block calculation rate distortion cost; Then they are compared, the minimum pattern of selection rate distortion cost is an optimum prediction mode.

In general, the large scale piecemeal is applicable to the smooth or stagnant zone of image; The small size piecemeal is suitable for image texture more complicated or the bigger zone of activity.And in the natural video frequency sequence, exist on the big quantity space static and the very little zone of activity on evenly smooth or the time domain.If can adopt some method to predict these zones, just can avoid the little minute complicated rate distortion costs of block mode to calculate, thereby reduce the scramble time greatly.The present invention is through selecting the plurality of pixels point of present image at random, and asks variance to handle to its luminance component, and the variance yields that obtains and pre-set threshold value are compared, and the result through comparison judges best inter-frame forecast mode fast.

Summary of the invention

Time, raising code efficiency for reducing inter prediction the present invention proposes a kind of H.264 inter-frame forecast mode fast selecting method based on the random pixel method.This method picked at random plurality of pixels point, and ask variance to handle to its luminance component, then variance yields and threshold value are compared, through optimum prediction mode between the quick judgment frame of comparative result.

In order to achieve the above object, the present invention adopts following technical scheme:

Step 1: select the plurality of pixels point of present image at random, and ask the variance yields δ of these pixel luminance components;

Step 2: set large scale threshold value TH ₁, small size threshold value TH ₂

Step 3: compare δ and threshold value TH ₁, TH ₂Relation, confirm best inter-frame forecast mode.

Wherein, in the said step 1: select the plurality of pixels point of present image at random, and ask the variance yields δ of these pixel luminance components, in the present invention, the selection of pixel is at random, and asks variance to handle to the luminance component of pixel.

Wherein, in the said step 2: set large scale threshold value TH ₁, small size threshold value TH ₂, in the present invention, large scale is set at 16 * 16 patterns; Small size is set at 8 * 8 patterns; The setting of threshold value along with prediction constantly carry out change, after the prediction of some frames finishes, the suitable correction threshold TH of the pairing variance yields δ of best inter-frame forecast mode that selects according to the front ₁, TH ₂

Wherein, in the said step 3: compare δ and threshold value TH ₁, TH ₂Relation, confirm best inter-frame forecast mode, in the present invention, if δ is smaller or equal to TH ₁, then selecting 16 * 16 patterns for use is best inter-frame forecast mode; If δ is more than or equal to TH ₂Then calculate and compare the rate distortion costs of 16 * 16 patterns and 8 * 8 patterns; If the rate distortion costs of 16 * 16 patterns is little; 16 * 16 patterns of selection are best inter-frame forecast mode, otherwise calculate the rate distortion costs of 8 * 8,8 * 4,4 * 8,4 * 4 patterns, and the minimum pattern of its rate distortion costs is best inter-frame forecast mode; If δ is greater than TH ₁Less than TH ₂, then calculate the rate distortion costs of 16 * 16,16 * 8,8 * 16 patterns, the minimum pattern of its rate distortion costs is best inter-frame forecast mode.Here work as δ greater than TH ₁The time, which kind of situation we all will calculate the rate distortion costs of 16 * 16 patterns, confirm optimum prediction mode with the rate distortion costs of other patterns again after relatively

Principle of the present invention: through calculating the variance yields select the pixel luminance component at random, and with itself and threshold value TH ₁, TH ₂Do comparison,, judge best inter-frame forecast mode fast according to result relatively.The advantage of this method is under the situation of not understanding the integral image textural characteristics, can judge best inter-frame forecast mode fast according to the variance yields of selecting the pixel luminance component at random, saves predicted time, improves code efficiency effectively.

Description of drawings

Fig. 1 is the basic structure flow chart of the inventive method;

Fig. 2 is the concrete realization flow figure of the inventive method.

Embodiment

The present invention is the new method that a kind of H.264 inter-frame forecast mode is selected.

Referring to accompanying drawing 1, main contents of the present invention are: select the plurality of pixels point of present image at random, and ask variance to handle to its luminance component, then with threshold value TH ₁, TH ₂Relatively, according to result relatively, judge best inter-frame forecast mode fast.This system of selection is selected suitable and threshold value is provided with under the suitable situation in the pixel number, can shorten the time of inter prediction, improves code efficiency.

For clearer elaboration the object of the invention, technical scheme and advantage,, be that the video sequence of CIF (352 * 288) is that example (supposing that its predictive mode is an inter-frame forecast mode) is done further detailed explanation with 300 frame formats below in conjunction with accompanying drawing.

Embodiment is seen Fig. 1 flow chart, and key step is following:

1, selects 200 pixels of current certain two field picture at random, at first extract these 200 pixel luma component values and ask its average, see formula (1), ask the variance yields of luminance component again, see formula (2).

$\mathrm{AVG}=\frac{1}{200}\underset{i=0}{\overset{200}{\mathrm{\Σ}}}{P}_i---\left(1\right)$

$\mathrm{\δ}=\frac{1}{200}\underset{i=0}{\overset{200}{\mathrm{\Σ}}}{(P_i-\mathrm{AVB})}^2---\left(2\right)$

AVG is the mean value of 200 pixel luminance components in above-mentioned two formulas, and δ is the variance yields of pixel luminance component, P _iBe the brightness value of i pixel, 200 numbers for selected pixels point can know that according to practical experience 200 pixels are enough to obtain enough forecasting accuracy.

2, the variance threshold values that large scale pattern 16 * 16 patterns is set is TH ₁, the variance threshold values of small size pattern 8 * 8 patterns is TH ₂, the TH here ₁, TH ₂Be to obtain at first, along with inter prediction constantly carries out, after every completion 100 frames prediction, to threshold value TH through practical experience ₁, TH ₂Revise, see formula (3), (4).

${\mathrm{TH}}_1=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\δ}}_i---\left(3\right)$

${\mathrm{TH}}_2=\frac{1}{h}\underset{j=1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{\δ}}_j---\left(4\right)$

N in the formula (3) is a number of selecting 16 * 16 patterns in the 100 frame inter predictions, δ _iPairing variance yields when selecting i 16 * 16 patterns for use, the h in the formula (4) is a number of selecting 8 * 8 patterns in the 100 frame inter predictions, δ _jPairing variance yields when selecting j 8 * 8 patterns for use, and n+h＜=100 are arranged.

3, with variance yields δ that obtains and threshold value TH ₁, TH ₂Relatively, select optimum prediction mode fast through comparative result.

(1) if δ smaller or equal to TH ₁, then selecting 16 * 16 patterns is best inter-frame forecast mode, other patterns are set to invalid mode.

(2) if δ more than or equal to TH ₂, then calculate the rate distortion costs of 16 * 16,8 * 8 patterns, be designated as RDC respectively ₁(Rate Distortion Cost), RDC ₂, if RDC ₁Smaller or equal to RDC ₂, then selecting 16 * 16 patterns is best inter-frame forecast mode, other patterns are set to invalid mode; If RDC ₁Greater than RDC ₂, then calculate the rate distortion costs of 8 * 8,4 * 8,8 * 4,4 * 4 patterns, selecting the minimum pattern of cost is best inter-frame forecast mode, other patterns are set to invalid mode.

(3) if δ greater than TH ₁Less than TH ₂, then calculate the rate distortion costs of 16 * 16,16 * 8,8 * 16 patterns, choosing the minimum pattern of rate distortion costs is best inter-frame forecast mode, other patterns are set to invalid mode.

The english abbreviation that occurs in the Figure of description, its implication is following:

RDC: rate distortion costs.

Claims (4)

1. the present invention proposes a kind of H.264 inter-frame forecast mode fast selecting method based on the random pixel method, the concrete steps of this method are following:

Step 1: select the plurality of pixels point of present image at random, and ask the variance yields δ of these pixel luminance components;

Step 2: set large scale threshold value TH ₁, small size threshold value TH ₂

Step 3: compare δ and threshold value TH ₁, TH ₂Relation, confirm best inter-frame forecast mode.

2. a kind of H.264 inter-frame forecast mode fast selecting method according to claim 1 based on the random pixel method; It is characterized in that: in the said step 1: the plurality of pixels point of selecting present image at random; And ask the variance yields δ of these pixel luminance components; In the present invention, the selection of pixel is at random, and asks variance to handle to the luminance component of pixel.

3. a kind of H.264 inter-frame forecast mode fast selecting method based on the random pixel method according to claim 1 is characterized in that: in the said step 2: set large scale threshold value TH ₁, small size threshold value TH ₂, in the present invention, large scale is set at 16 * 16 patterns, and small size is set at 8 * 8 patterns, and threshold value is variable, after the prediction of some frames finishes, according to the suitable correction threshold TH of the pairing variance yields δ of best inter-frame forecast mode of front selection ₁, TH ₂

4. a kind of H.264 inter-frame forecast mode fast selecting method based on the random pixel method according to claim 1 is characterized in that: in the said step 3: compare δ and threshold value TH ₁, TH ₂Relation, confirm best inter-frame forecast mode, in the present invention, if δ is smaller or equal to TH ₁, then selecting 16 * 16 patterns for use is best inter-frame forecast mode; If δ is more than or equal to TH ₂Then calculate and compare the rate distortion costs of 16 * 16 patterns and 8 * 8 patterns; If the rate distortion costs of 16 * 16 patterns is little; Then selecting 16 * 16 patterns for use is best inter-frame forecast mode, otherwise calculates the rate distortion costs of 8 * 8,8 * 4,4 * 8,4 * 4 patterns, and the minimum pattern of selection rate distortion cost is best inter-frame forecast mode; If δ is greater than TH ₁Less than TH ₂, then calculate the rate distortion costs of 16 * 16,16 * 8,8 * 16 patterns, the minimum pattern of selection rate distortion cost is best inter-frame forecast mode.

Images