CN1738431A - Frame field self-adaptive detection method - Google Patents

Abstract

The invention provides a frame-field self-adaptive detecting method, which calculates the characteristic value C1 of input image according to the input image; finding the detecting threshold Tn of present frame according to the first N frames of input image; when C1<Tn, and C1<1, the frame uses the encoding mode of frame encoding, or else, the frame uses the encoding code of field encoding. The inventive frame-field self-adaptive detecting method has high accuracy of encoding and high efficiency.

Description

A kind of frame field self-adaptive detection method

Technical field

The present invention relates to a kind of frame field self-adaptive detection method, relate in particular to the frame field self-adaptive detection method of a kind of image level for Interlace (interlacing) sequence in the audio/video encoding/decoding technical field in signal processing.

Background technology

Interlacing scan (Interlace scan) is a kind of scan format of often using in television system, it is that a two field picture is decomposed into two, wherein the field, top only comprises the odd-numbered line of image, field, the end only comprises the even number line of image, it relies on the eye storage characteristic of human eye and some characteristics of display, the scan line of two fields is appeared to be interleaved in together, become a complete image, the great advantage of this scan format is to realize higher refresh rate with half the data volume of lining by line scan, make picture not have the sensation of flicker, but it also has a very big shortcoming, be exactly that amount of exercise when image is when big, when two field picture shows as a two field picture since the time difference between two have sawtooth between two row of image and occur.

In recent years a collection of digital audio/video encoding and decoding standard of Chu Xianing, representative have international standard MPEG-4, an AVC H.264/MPEG-4, the autonomous standard A VS that formulates of China, WM9 that Microsoft releases or the like, these standards have all related to the processing to the Interlace sequence.

In MPEG-2 and MPEG-4 standard, characteristics according to the Interlace sequence, be provided with " frame coding " and " coding " two kinds of patterns, " frame coding " is meant the pattern that encoder is encoded as the elementary cell of encoder with the frame macro block of a frame of two occasions one-tenth, and " coding " is meant the pattern that encoder is encoded as the elementary cell of encoder with every field macro block.Encoder is encoded to every width of cloth image according to pre-set pattern, itself does not have the function according to the adaptive selection coding mode of the characteristics of sequence, may cause an original image encoded that is fit to be encoded like this by frame, perhaps be fit to the frame image encoded originally and encoded, reduced coding quality with regard to making because of the selection mistake of coding mode like this by the field; In H.264/MPEG-4 AVC and AVS, encoder carries out a frame coding and a coding at first respectively to vision signal, then according to both coding cost, a still coding of frame coding is adopted in decision, that is to say that H.264/MPEG-4 AVC and AVS have adopted the method for compiling twice to realize coding to the frame field adaptive of the image level of interlace sequence, code efficiency has reduced half like this, and therefore this method is difficult to be applied to the demanding occasion of real-time.

Summary of the invention

A kind of frame field self-adaptive detection method provided by the invention, accuracy rate height, efficient height.

In order to achieve the above object, the invention provides a kind of frame field self-adaptive detection method, may further comprise the steps:

The characteristic value C of step 1, calculating input image ₁:

If the size of input picture is Width pixel * Height pixel, and pixel (x, brightness value y) be I (x y), then has:

$C_{\mathrm{frame}\_\mathrm{top}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,y)-I(x,y+1)|---\left(1\right)$

$C_{\mathrm{frame}\_\mathrm{bot}}=\underset{y=\frac{\mathrm{<figure-callout\; id="2"\; label="Height"\; filenames="A20051002952600111.png"\; state="\{\{state\}\}">Height</figure-callout>}}{2}}{\overset{\mathrm{Height}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,y)-I(x,y+1)|---\left(2\right)$

$C_{\mathrm{field}\_\mathrm{top}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-2}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,2y)-I(x,2y+2)|---\left(3\right)$

$C_{\mathrm{field}\_\mathrm{bot}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-2}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,2y+1)-I(x,2y+3)|---\left(4\right)$

So, the characteristic value C of input picture ₁Value be:

$C_1=\frac{C_{\mathrm{frame}\_\mathrm{top}}+C_{\mathrm{frame}\_\mathrm{bot}}}{C_{\mathrm{field}\_\mathrm{top}}+C_{\mathrm{field}\_\mathrm{bot}}}---\left(5\right)$

For the Interlace sequence, the field, top of its image and the time of origin of field, the end have certain time interval, and when the part of strenuous exercise was arranged in the image, two when showing as frame coding, the violent part of moving just had tangible sawtooth;

The characteristic value C of the input picture that calculates ₁More little, then the amount of exercise of this two field picture is more little, illustrates that this two field picture is more suitable in frame encoding mode; On the contrary, if the characteristic value C of input picture ₁Big more, then the amount of exercise of this two field picture is big more, illustrates that this picture frame is more suitable in the field coding mode;

Step 2, the final detection threshold T that determines current n frame according to the preceding N two field picture and the frame type of present frame of input _nWhen n＜N, detection threshold can be brought in constant renewal in according to the pixel data of the new input of every frame, obtain the benchmaring thresholding T of current n frame _NjWhen; During n＞=N, the benchmaring thresholding is T _NjDetermine the final detection threshold T of this frame then according to the frame type of this frame _nThis step comprises step by step following:

Step 2.1, current incoming frame is set is the n frame, whether judges n greater than N, if, direct execution in step 2.4 then; If not, then execution in step 2.2 and step 2.3 successively;

The characteristic value C of step 2.2, calculating present frame ₂, the angle of each macro block that it is cut apart from current frame image has illustrated the momental size of current frame image, it comprises step by step following:

Step 2.2.1, two counting variable: N of setting _Field=0, N _Frame=0;

Step 2.2.2, for the image of present frame, be divided into the macro block of several M pixel * M pixels, wherein, M ∈ [4,8,16,32,64], for each macro block, calculate its top auto-variance V _TopAuto-variance V with field, the end _Bot

$V_{\mathrm{top}}=\underset{y=0}{\overset{\frac{M}{2}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{M-1}{\mathrm{\&Sigma;}}}{(I(x,2y)-E)}^2---\left(6\right)$

$V_{\mathrm{bot}}=\underset{y=0}{\overset{\frac{M}{2}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{M-1}{\mathrm{\&Sigma;}}}{(I(x,2y+1)-E)}^2---\left(7\right)$

Wherein, E represents the mean value of brightness of all pixels of this macro block;

Step 2.2.3, for the V of each macro block _TopAnd V _Bot, calculate A=V _Top/ V _Bot, if $A\&NotElement;[M{T}_1,M{T}_2],$ Counting variable N then _FieldValue add 1; If A ∈ [MT ₁, MT ₂], counting variable N then _FrameValue add 1, wherein, 1.0＜MT ₁＜1.5,0.5＜MT ₂＜1.0;

Step 2.2.4, computation of characteristic values C ₂=N _Frame/ N _Field, C ₂Value big more, the suitable more employing frame of this two field picture coding then is described;

Step 2.3, determine detection threshold, it comprises step by step following:

Step 2.3.1, computed image present frame, i.e. the initial examination and measurement thresholding of n frame:

Work as C ₂〉=Mod ₁During * M, $T_f\left(n\right)=\frac{\mathrm{Cof}\&times;{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20051002952600111.png"\; state="\{\{state\}\}">C</figure-callout>}}_2\&times;C_1}{4}$ , wherein, 0.9＜Cof＜1.0,0.25＜Mod ₁＜0.5;

Work as C ₂〉=Mod ₂During * M, $T_f\left(n\right)=\frac{\mathrm{Cof}\&times;{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20051002952600111.png"\; state="\{\{state\}\}">C</figure-callout>}}_2\&times;C_1}{4},$ Wherein, 0.8＜Cof＜0.9; 0.2＜Mod ₂＜0.4;

Work as C ₂＜Mod ₂During * M, $T_f\left(n\right)=\frac{{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20051002952600111.png"\; state="\{\{state\}\}">C</figure-callout>}}_2\&times;C_1}{4},$ Wherein, 0.2＜Mod ₂＜0.4;

Step 2.3.2, consider the correlation of front n-1 frame and n frame, the benchmaring thresholding of n frame is: $T_{\mathrm{nj}}=\frac{1}{n}\underset{k=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{T}_f\left(k\right);$

Step 2.3.3, according to present frame, the i.e. frame type of n frame, determine the final detection threshold of this frame: according to the experience result of frame field adaptive coding among H.264/MPEG-4 AVC and the AVS, because figure (I frame), prognostic chart (P frame) are compared with bi-directional predicted figure (B frame) in the frame, be more suitable in the frame coding, therefore for I frame and P frame, detection threshold is less, for the B frame, detection threshold is bigger, so;

For I frame or P frame, final detection threshold is: T _n=T _Nj

For the B frame, final detection threshold is: T _n=Coe * T _Nj, wherein, 1.0＜Coe＜2.0;

Step 2.4, because the number of image frames n of input greater than N, represents that the frame field detection threshold of this sequence is basicly stable, at this moment can be directly determine the detection threshold of this frame, that is: according to the detection threshold of the frame type of present frame and preceding N frame

For I frame or P frame, order _n=T _Nj

For the B frame, make T _n=Coe * T _Nj, wherein, 1.0＜Coe＜2.0;

Step 3, image level frame field adaptive detect:

If C ₁＜T _n, and C ₁＜1, then this frame adopts the coding mode of frame coding, otherwise this frame adopts a coding mode of coding.

Frame field self-adaptive detection method provided by the invention, from whole and local two aspects, momental size to every frame has been carried out quantitative estimation, and the preceding N frame of each sequence integrated consideration, progressively determine the thresholding of the frame field detection of this sequence, according to this thresholding, pre-determine the coding mode of each frame, so just can both guarantee the correctness that higher coding mode is selected, improve coding quality, avoided again H.264/MPEG-4AVC with AVS in be cost to sacrifice code efficiency, to improve coding quality.

H.264/MPEG-4AVC a kind of frame field self-adaptive detection method that this invention provides can be used for and encoders such as AVS, these encoders can be used the method among the present invention, before coding, each two field picture to the interlace sequence carries out the frame field adaptive detection, according to the coding mode of the every frame that obtains image is encoded then, can not improve encoder complexity like this, not influence on the basis of coding rate, improve coding quality.

Description of drawings

Fig. 1 is the flow chart of frame field self-adaptive detection method provided by the invention;

Fig. 2 is the flow chart of definite detection threshold of frame field self-adaptive detection method provided by the invention.

Embodiment

Followingly specify preferred forms of the present invention according to Fig. 1～Fig. 2:

As shown in Figure 1 and Figure 2, the invention provides a kind of frame field self-adaptive detection method, may further comprise the steps:

The characteristic value C of step 1, calculating input image ₁:

If the size of input picture is Width pixel * Height pixel, and pixel (x, brightness value y) be I (x y), then has:

$C_{\mathrm{frame}\_\mathrm{top}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,y)-I(x,y+1)|---\left(1\right)$

$C_{\mathrm{frame}\_\mathrm{bot}}=\underset{y=\frac{\mathrm{<figure-callout\; id="2"\; label="Height"\; filenames="A20051002952600111.png"\; state="\{\{state\}\}">Height</figure-callout>}}{2}}{\overset{\mathrm{Height}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,y)-I(x,y+1)|---\left(2\right)$

$C_{\mathrm{field}\_\mathrm{top}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-2}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,2y)-I(x,2y+2)|---\left(3\right)$

$C_{\mathrm{field}\_\mathrm{bot}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-2}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,2y+1)-I(x,2y+3)|---\left(4\right)$

So, the characteristic value C of input picture ₁Value be:

$C_1=\frac{C_{\mathrm{frame}\_\mathrm{top}}+C_{\mathrm{frame}\_\mathrm{bot}}}{C_{\mathrm{field}\_\mathrm{top}}+C_{\mathrm{field}\_\mathrm{bot}}}---\left(5\right)$

For the Interlace sequence, the field, top of its image and the time of origin of field, the end have certain time interval, and when the part of strenuous exercise was arranged in the image, two when showing as frame coding, the violent part of moving just had tangible sawtooth;

The characteristic value C of the input picture that calculates ₁More little, then the amount of exercise of this two field picture is more little, illustrates that this two field picture is more suitable in frame encoding mode; On the contrary, if the characteristic value C of input picture ₁Big more, then the amount of exercise of this two field picture is big more, illustrates that this picture frame is more suitable in the field coding mode;

Step 2, the final detection threshold T that determines current n frame according to the preceding N two field picture and the frame type of present frame of input _nWhen n＜N, detection threshold can be brought in constant renewal in according to the pixel data of the new input of every frame, obtain the benchmaring thresholding T of current n frame _NjWhen n＞=N, the benchmaring thresholding is T _NjDetermine the final detection threshold T of this frame then according to the frame type of this frame _nThis step comprises step by step following:

Step 2.1, current incoming frame is set is the n frame, whether judges n greater than N, if, direct execution in step 2.4 then; If not, then execution in step 2.2 and step 2.3 successively;

The characteristic value C of step 2.2, calculating present frame ₂, the angle of each macro block that it is cut apart from current frame image has illustrated the momental size of current frame image, it comprises step by step following:

Step 2.2.1, two counting variable: N of setting _Field=0, N _Frame=0;

Step 2.2.2, for the image of present frame, may be partitioned into the macro block of several 16 pixels * 16 pixels, for each macro block, calculate its top auto-variance V _TopAuto-variance V with field, the end _Bot

$V_{\mathrm{top}}=\underset{y=0}{\mathrm{\&Sigma;}}\underset{x=0}{\overset{15}{\mathrm{\&Sigma;}}}{(I(x,2y)-E)}^2---\left(6\right)$

$V_{\mathrm{bot}}=\underset{y=0}{\mathrm{\&Sigma;}}\underset{x=0}{\overset{15}{\mathrm{\&Sigma;}}}{(I(x,2y+1)-E)}^2---\left(7\right)$

Wherein, E represents the mean value of brightness of all pixels of this macro block;

Step 2.2.3, for the V of each macro block _TopAnd V _Bot, calculate A=V _Top/ V _Bot, if $A\&NotElement;[{\mathrm{MT}}_1,{\mathrm{MT}}_2],$ Counting variable N then _FieldValue add 1; If A ∈ [MT ₁, MT ₂], counting variable N then _FrameValue add 1, wherein, 1.0＜MT ₁＜1.5,0.5＜MT ₂＜1.0;

Step 2.2.4, computation of characteristic values C ₂=N _Frame/ N _Field, C ₂Value big more, the suitable more employing frame of this two field picture coding then is described;

Step 2.3, determine detection threshold, it comprises step by step following:

Step 2.3.1, computed image present frame, i.e. the initial examination and measurement thresholding of n frame:

Work as C ₂〉=Mod ₁During * M, $T_f\left(n\right)=\frac{\mathrm{Cof}\&times;{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20051002952600111.png"\; state="\{\{state\}\}">C</figure-callout>}}_2\&times;C_1}{4},$ Wherein, 0.9＜Cof＜1.0,0.25＜Mod ₁＜0.5;

Work as C ₂〉=Mod ₂During * M, $T_f\left(n\right)=\frac{\mathrm{Cof}\&times;{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20051002952600111.png"\; state="\{\{state\}\}">C</figure-callout>}}_2\&times;C_1}{4},$ Wherein, 0.8＜Cof＜0.9; 0.2＜Mod ₂＜0.4;

Work as C ₂＜Mod ₂During * M, $T_f\left(n\right)=\frac{{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="A20051002952600111.png"\; state="\{\{state\}\}">C</figure-callout>}}_2\&times;C_1}{4},$ Wherein, 0.2＜Mod ₂＜0.4; Step 2.3.2, consider the correlation of front n-1 frame and n frame, the benchmaring thresholding of n frame is: $T_{\mathrm{nj}}=\frac{1}{n}\underset{k=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{T}_f\left(k\right);$

Step 2.3.3, according to present frame, the i.e. frame type of n frame, determine the final detection threshold of this frame: according to the experience result of frame field adaptive coding among H.264/MPEG-4 AVC and the AVS, figure (I frame), prognostic chart (P frame) are compared with bi-directional predicted figure (B frame) in the frame, be more suitable in the frame coding, therefore for I frame and P frame, detection threshold is less, bigger for B frame detection threshold, so:

For I frame or P frame, final detection threshold is: T _n=T _Nj

For the B frame, final detection threshold is: T _n=Coe * T _Nj, wherein, 1.0＜Coe＜2.0;

Step 2.4, because the number of image frames n of input greater than N, represents that the frame field detection threshold of this sequence is basicly stable, at this moment can be directly determine the detection threshold of this frame, that is: according to the detection threshold of the frame type of present frame and preceding N frame

For I frame or P frame, make T _n=T _Nj

For the B frame, make T _n=Coe * T _Nj, wherein, 1.0＜Coe＜2.0;

Step 3, image level frame field adaptive detect:

If C ₁＜T _n, and C ₁＜1, then this frame adopts the coding mode of frame coding, otherwise this frame adopts a coding mode of coding.

Frame field self-adaptive detection method provided by the invention, from whole and local two aspects, momental size to every frame has been carried out quantitative estimation, and the preceding N frame of each sequence integrated consideration, progressively determine the thresholding of the frame field detection of this sequence, according to this thresholding, pre-determine the coding mode of each frame, so just can both guarantee the correctness that higher coding mode is selected, improve coding quality, avoided again H.264/MPEG-4AVC with AVS in be cost to sacrifice code efficiency, to improve coding quality.

H.264/MPEG-4AVC a kind of frame field self-adaptive detection method that this invention provides can be used for and encoders such as AVS, these encoders can be used the method among the present invention, before coding, each two field picture to the interlace sequence carries out the frame field adaptive detection, according to the coding mode of the every frame that obtains image is encoded then, can not improve encoder complexity like this, not influence on the basis of coding rate, improve coding quality.

Claims (1)

1. a frame field self-adaptive detection method is characterised in that, may further comprise the steps:

The characteristic value C of step 1, calculating input image ₁:

If the size of input picture is Width pixel * Height pixel, and pixel (x, brightness value y) be I (x y), then has:

$\begin{array}{c}{C}_{\mathrm{frame}\_\mathrm{top}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,y)-I(x,y+1)|;\\ C_{\mathrm{frame}\_\mathrm{bot}}=\underset{y=\frac{\mathrm{Height}}{2}}{\overset{\mathrm{Height}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,y)-I(x,y+1)|;\\ C_{\mathrm{field}\_\mathrm{top}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-2}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,2y)-I(x,2y+2)|;\\ C_{\mathrm{field}\_\mathrm{bot}}=\underset{y=0}{\overset{\frac{\mathrm{Height}}{2}-2}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{\mathrm{Width}-1}{\mathrm{\&Sigma;}}}|I(x,2y+1)-I(x,2y+3)|;\end{array}$

So, the characteristic value C of input picture ₁Value be: $C_1=\frac{C_{\mathrm{frame}\_\mathrm{top}}+C_{\mathrm{frame}\_\mathrm{bot}}}{C_{\mathrm{field}\_\mathrm{top}}+C_{\mathrm{field}\_\mathrm{bot}}};$

Step 2, the detection threshold T that determines current n frame according to the preceding N two field picture and the frame type of present frame of input _nComprise step by step following:

Step 2.1, current incoming frame is set is the n frame, whether judges n greater than N, if, direct execution in step 2.4 then; If not, then execution in step 2.2 and step 2.3 successively;

The characteristic value C of step 2.2, calculating present frame ₂, it comprises step by step following:

Step 2.2.1, two counting variable: N of setting _Field=0, N _Frame=0;

Step 2.2.2, for the image of present frame, be divided into the macro block of several M pixel * M pixels, wherein, M ∈ [4,8,16,32,64]; For each macro block, calculate the auto-variance V of its field, top _TopAuto-variance V with field, the end _Bot

$\begin{array}{c}{V}_{\mathrm{top}}=\underset{y=0}{\overset{\frac{M}{2}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{M-1}{\mathrm{\&Sigma;}}}{(I(x,2y)-E)}^2;\\ V_{\mathrm{bot}}=\underset{y=0}{\overset{\frac{M}{2}-1}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{M-1}{\mathrm{\&Sigma;}}}{(I(x,2y+1)-E)}^2;\end{array}$

Wherein, E represents the mean value of brightness of all pixels of this macro block;

Step 2.2.3, for the V of each macro block _TopAnd V _Bot, calculate A=V _Top/ V _Bot, if $A\&NotElement;[M{T}_1,M{T}_2]$ , counting variable N then _FieldValue add 1; If A ∈ [MT ₁, MT ₂], counting variable N then _FrameValue add 1, wherein, 1.0＜MT ₁＜1.5,0.5＜MT ₂＜1.0;

Step 2.2.4, computation of characteristic values C ₂=N _Frame/ N _Field

Step 2.3, determine detection threshold, it comprises step by step following:

Step 2.3.1, computed image present frame, i.e. the initial examination and measurement thresholding of n frame:

Work as C ₂〉=Mod ₁During * M, $T_f\left(n\right)=\frac{\mathrm{Cof}\&times;C_2\&times;C_1}{4},$ Wherein, 0.9＜Cof＜1.0,0.25＜Mod ₁＜0.5;

Work as C ₂〉=Mod ₂During * M, $T_f\left(n\right)=\frac{\mathrm{Cof}\&times;C_2\&times;C_1}{4}$ Wherein, 0.8＜Cof＜0.9; 0.2＜Mod ₂＜0.4;

Work as C ₂＜Mod ₂During * M, $T_f\left(n\right)=\frac{C_2\&times;C_1}{4},$ Wherein, 0.2＜Mod ₂＜0.4;

Step 2.3.2, present frame, promptly the benchmaring thresholding of n frame is: $T_{\mathrm{nj}}=\frac{1}{n}\underset{k=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{T}_f\left(k\right);$

Step 2.3.3, according to present frame, i.e. the frame type of n frame, determine the final detection threshold of this frame:

For figure I frame or prognostic chart P frame in the frame, final detection threshold is: T _n=T _Nj

For bi-directional predicted figure B frame, final detection threshold is: T _n=Coe * T _Nj, wherein, 1.0＜Coe＜2.0;

The number of image frames n of step 2.4, input then has greater than N:

For figure I frame or prognostic chart P frame in the frame, final detection threshold is: T _n=T _Nj

For bi-directional predicted figure B frame, final detection threshold is: T _n=Coe * T _Nj, wherein, 1.0＜Coe＜2.0;

Step 3, image level frame field adaptive detect: if C ₁＜T _n, and C ₁＜1, then this frame adopts the coding mode of frame coding, otherwise this frame adopts a coding mode of coding.

Images