CN1694494A - Detection method of 3:2 pull-down film source video - Google Patents

Abstract

This invention relates to a 3:2 pull-down film source video detection method, which carries out computation, correction, match and judgment to a receive video frequency to test if it's a 3:2 pull-down film source video frequency with a field difference measurement module, a field difference measurement module, a field difference correct module, a field label prediction module, a field label match module and a mode judgment module, which can effectively test the film source video frequency in a mixed-edited video frequency and switches to the de-interleave mode quickly and accurately when changes happen to the video.

Description

The detection method of the drop-down film source video of a kind of 3:2

Technical field

The invention belongs to and calculate video, television scanning format conversion technical field, relate to and detect whether input video is the method for the drop-down film source video of 3:2 in a kind of digital television system, so which kind of method decision takes go interlacing to handle to interlaced video according to testing result.

Background technology

Because vision bandwidth is directly proportional with the line number of scan line, for bandwidth is controlled within the acceptable range, the simulated television technology in past has adopted interleaved mode, but this is cost with the decrease in image quality, especially the abundant or violent zone of moving in the image vertical detail has significantly the flaws such as shake and flicker of creeping, go.Because there are a lot of intrinsic defectives in interlaced scan tv, along with the raising of user to the picture quality requirement, increasing television manufacturer has been released progressive-scan tv, but in existing television system, transmission all be interlace signal, need go interlacing to handle at receiving terminal and could show progressive, in addition, even following television system is based on progressive,, also need the interleaved video data of a large amount of preciousnesses is converted to progressive format at transmitting terminal.

In the interlaced video of existing television system transmission, there is quite a few to be transformed by progressive, be called the film source video, comprise most of film, part TV play, advertisement, documentary film and with the cartoon of computer photograph etc.

Film source video refer initially to is that the cinefilms of 24 hardwood/seconds transforms the interlaced video obtain by the hardwood rate.In TSC-system, drop-down by 3:2, be converted into 60/second interlaced video signal.As shown in Figure 1, the 3:2 pulldown method is an one-period with 4 frame films, and first frame repeats once strange field 01 after being split as a strange field 01 and an idol E1 again; Second frame is split as an idol E2 and strange field 02; After the 3rd frame is split as an idol E3 and strange field 03, repeat an once idol E3 again; The 4th frame then is split as strange 04 and an idol E4.So not only guaranteed the basic principle that parity field is adjacent, also field frequency has been transformed to 60/second.

For the film source video,,, and, just can obtain not having the original progressive video of flaw with after its merging as long as can detect strange and even field signal from same frame because its data source is line by line.

But mixed transport film source video and normal interlaced vision signal in the common television system, and this just need effectively detect the video type of mixed editorial, can correctly switch interlaced mode rapidly when video type changes.There is following shortcoming usually in the detection method of existing film source video:

1. make mistakes easily in the junction of different camera lenses, scene;

2. can not improve speed and the reliability that enters film mode simultaneously;

3. static or the very small video of motion are judged by accident easily;

4. threshold value depends on video source, but because the complexity of rule, is difficult to realize the self adaptation of threshold value.

Summary of the invention

In order to improve the defective that exists in the existing film source video detecting method, the objective of the invention is to, a kind of detection method that can distinguish drop-down film source video of 3:2 and normal interlaced video is provided.In the video type of mixed editorial, especially when video type changed, this method can detect video type exactly, and interlaced mode is gone in correct switching fast.

The technical solution adopted in the present invention is, the detection method of the drop-down film source video of 3:2, by comprising a difference metric module, a difference correcting module, a label generation module, a label prediction module, a label matching module and pattern judge module the video type that receives is detected, to judge whether it is the drop-down film source video of 3:2, this method may further comprise the steps

The data input difference metric module that to work as first like parity field of front court and its front is calculated two poor in the difference metric module on the scene, the number of statistical difference dissimilarity, and export the poor correcting module of showing up as the field diversity factor;

Field difference correcting module is according to the pattern judged result of previous field, if not film mode, then the field diversity factor that receives is not detected and revises, if film mode then detects the field diversity factor, when it is the anomalous field difference when detection, it is revised, together as current corrected fields difference vector, and the output label generation module of showing up outputs to the pattern judge module with the correction marking signal of generation through field diversity factor revised and preceding four corrected fields diversity factor;

Field label generation module carries out the detection of least member to the corrected fields difference vector that receives, and with the position of least member as this actual label output label matching module of showing up, if the ratio of the least member of corrected fields difference vector and other four elements is all less than a threshold value, then label sign in field is set to 0, otherwise field label sign is set to 1, and field label sign is outputed to the pattern judge module;

Field label prediction module reads last one actual label from memory, try to achieve the prediction label when the front court, and exports the label matching module of showing up;

In the label matching module on the scene, prediction label and the actual label when the front court that receive are compared, generate mode flags signal, read last one field mode conceptual vector from memory when the front court, generate this field mode conceptual vector, and output to the pattern judge module;

The pattern judge module receives revises marking signal, a label sign and field mode conceptual vector, be 1 if revise marking signal with a label sign, and the field mode conceptual vector is omnidirectional's amount, then video is the film source video, the pattern judged result is 1, otherwise the pattern judged result is 0, the pattern judged result is outputed to the interlaced mode control system, and control goes interlaced mode to enter film mode or general mode.

Characteristics of the present invention also are, in the poor correcting module on the scene, the method that the field diversity factor is detected is, obtain a poor correction value according to the corrected fields difference vector of previous field, if follow a corresponding threshold value of difference correction value from the field diversity factor of field difference metric module input greater than one, think that then this difference is that anomalous field is poor.

Another characteristics of the present invention are, in the poor correcting module on the scene, to the modification method of anomalous field difference are, determine whether that according to the camera lens sign needs revise the anomalous field difference, if then use difference correction value replaces a diversity factor.

The invention has the beneficial effects as follows: can effectively distinguish the film source video type in the video of mixed editorial, when changing, video type can correctly switch interlaced mode rapidly, still can carry out reliable detection in camera lens switching place, recognition accuracy to static and dynamic menu is all very high, the part threshold value has realized self adaptation, and hardware is realized simply, cost is low.

Description of drawings

Fig. 1 is the schematic diagram of 3:2 pulldown method;

Fig. 2 is a detection method schematic flow sheet of the present invention;

Fig. 3 is a schematic diagram of difference adjusting module;

Fig. 4 is that schematic diagram is handled in the interlacing of going of film source video.

Embodiment

The present invention is described in detail below in conjunction with accompanying drawing.

As shown in Figure 1, the 3:2 pulldown method is an one-period with 4 frame films, and first frame repeats once strange field 01 after being split as a strange field 01 and an idol E1 again; Second frame is split as an idol E2 and strange field 02; After the 3rd frame is split as an idol E3 and strange field 03, repeat an once idol E3 again; The 4th frame then is split as strange 04 and an idol E4.So not only guaranteed the basic principle that parity field is adjacent, also field frequency has been transformed to 60/second.

Referring to Fig. 2, detection method of the present invention detects by comprising a difference metric module 1, a difference correcting module 2, a label generation module 3, a label prediction module 4, a label matching module 5 and 6 pairs of video type that receive of pattern judge module, specifically may further comprise the steps

1. the data input difference metric module 1 that will work as first like parity field of front court and its front is calculated two poor in the difference metric module 1 on the scene, the number of statistical difference dissimilarity, and export the poor correcting module 2 of showing up as the field diversity factor;

To n field signal f _n, use f _n(i j) represents that it is capable at i, the brightness value of j row, and first like parity field is f before it _N-2, ask f earlier _nAnd f _N-2Antipode dif _n

dif _n(i，j)＝|f _n(i，j)-f _n-2(i，j)|

Utilize fixed threshold thDif to antipode dif then _nCarry out binaryzation, obtain discrepancy sign difP _n(i, j), discrepancy sign difP _n(i j) is 1 expression two field picture (i, j) locational some pixel value difference, discrepancy sign difP _n(i, j) be 0 the expression two field picture (i, j) a locational pixel value is identical, their fine difference is just caused by noise.Discrepancy sign difP _n(i j) calculates according to following formula

${\mathrm{difP}}_n(i,j)=\left\{\begin{array}{ccc}1& \mathrm{if}& \mathrm{di}{f}_n(i,j)\&GreaterEqual;\mathrm{<figure-callout\; id="0"\; label="thDif"\; filenames="A20051004277800151.png"\; state="\{\{state\}\}">thDif</figure-callout>}\\ 0& \mathrm{if}& \mathrm{di}{f}_n(i,j)<\mathrm{thDif}\end{array}\right.$

Use f _nAnd f _N-2The number difC of two middle discrepancys _nRepresent f _nThe forward direction same sex field diversity factor of field is calculated according to following formula

${\mathrm{difC}}_n=\underset{i,j}{\mathrm{\&Sigma;}}{\mathrm{difP}}_n(i,j).$

2. a difference correcting module 2, does not then detect and revises the field diversity factor that receives if not film mode according to the pattern judged result of previous field, if film mode then detects the anomalous field diversity factor, and according to it is optionally revised.Together as current corrected fields difference vector, and the output label generation module 3 of showing up outputs to pattern judge module 6 with the correction marking signal of generation through field diversity factor revised and preceding four corrected fields diversity factor;

1) detection of anomalous field diversity factor

Represent with abnormality mark abnorm whether a diversity factor is unusual, calculate abnorm according to following formula

$\mathrm{abnorm}=\left\{\begin{array}{ccc}1& \mathrm{if}& \mathrm{dif}{C}_n\&GreaterEqual;\mathrm{co}\&times;\mathrm{<figure-callout\; id="0"\; label="difModB"\; filenames="A20051004277800151.png"\; state="\{\{state\}\}">difModB</figure-callout>}\\ 0& \mathrm{else}& \end{array}\right.$

Wherein, abnormality mark abnorm=1 represents that a diversity factor is unusual, and the field that a diversity factor is unusual is called anomalous field; Abnormality mark abnorm=0 represents that a diversity factor is normal.Co (1＜co＜2) is a coefficient of setting in advance.DifModB is the reference field diversity factor, and it is calculated as follows shown in the part.

2) calculating of corrected fields diversity factor and reference field diversity factor

If last one testing result is a film mode, and the number abnDif of anomalous field is 0, then calculates little diversity factor correction value difModS and reference field diversity factor difModB according to following formula:

difModS＝min(M _n-1)

$\mathrm{difModB}=\frac{(\underset{m=1}{\overset{5}{\mathrm{\&Sigma;}}}{M}_{n-1}\left(m\right)-\mathrm{difModS})}{4}$

Use mDifC _nExpression corrected fields diversity factor is calculated mDifC according to following formula _n:

Wherein, PN _nIt is prediction label when the front court.

Corrected fields difference vector M _N-1The middle f that preserves _N-5, f _N-4, f _N-3And f _N-1Five revised forward direction same sex field diversity factor mDifC _N-5, mDifC _N-4, mDifC _N-3, mDifC _N-2And mDifC _N-1, promptly

M _n-1＝[mDifC _n-5，mDifC _n-4，mDifC _n-3，mDifC _n-2，mDifC _n-1]

3), revise the generation of sign

Reason according to the anomalous field diversity factor produces can be divided into two classes with the anomalous field diversity factor in the video: the anomalous field diversity factor that a class occurs when to be video type by the film source video become non-film source video; Another kind of is the inner anomalous field diversity factor that occurs of film source video, for example switches the anomalous field diversity factor that causes by camera lens.To treat two class anomalous field diversity factoies in order distinguishing,, to withdraw from film mode fast and go interlacing, and when the second class anomalous field diversity factor occurring, remain on the film mode interlace-removing method, adopt method as shown in Figure 3 so that when first kind anomalous field diversity factor occurring.

The initial value of revising sign flagMod is 1, and anomalous field number abnDif represents the number of the anomalous field that this is before continuous, and its initial value is 0.When anomalous field number abnDif=0, if abnormality mark abnorm=1, then anomalous field number abnDif adds 1; When anomalous field number abnDif=1, if abnormality mark abnorm=1, then anomalous field number abnDif adds 1, otherwise makes anomalous field number abnDif=0; When anomalous field number abnDif=2 or abnDif=3, if abnormality mark abnorm=1, then anomalous field number abnDif adds 1, otherwise makes anomalous field number abnDif=0; When anomalous field number abnDif=4, make anomalous field number abnDif=0, if abnormality mark abnorm=1, then sign flagMod=0 is revised in order, otherwise order prediction label PV=5, and to make preceding four of the field mode conceptual vector be 1.

3. 3 pairs of corrected fields difference vectors that receive of label generation module carry out the detection of least member, and with the position of least member as this actual label output label matching module 5 of showing up, if the ratio of the least member of corrected fields difference vector and other four elements is all less than a threshold value, then label sign in field is set to 0, otherwise field label sign is set to 1, and field label sign is outputed to pattern judge module 6;

Field label generation method

According to corrected fields difference vector M _nCalculate actual label TN when the front court _n

${\mathrm{TN}}_n=\left\{\begin{array}{ccc}1& \mathrm{if}& \min \left(M_n\right)=\mathrm{dif}{C}_{n-4}\\ 2& \mathrm{if}& \min \left(M_n\right)=\mathrm{dif}{C}_{n-3}\\ 3& \mathrm{if}& \min \left(M_n\right)=\mathrm{dif}{C}_{n-2}\\ 4& \mathrm{if}& \min \left(M_n\right)=\mathrm{dif}{C}_{n-1}\\ 5& \mathrm{if}& \min \left(M_n\right)=\mathrm{dif}{C}_n\end{array}\right.$

Corrected fields difference vector M _nLeast member be min (M _n), sub-minimum is sec min (M _n), ask a label sign flagtn according to following formula _n, and it is outputed to the pattern judge module.

4. a label prediction module 4 reads last one actual label from memory, tries to achieve the prediction label when the front court, and exports the label matching module 5 of showing up;

Field label Forecasting Methodology

This module is according to the actual label TN of previous field _N-1Generate prediction label PN when the front court _n

${\mathrm{PN}}_n=\left\{\begin{array}{ccc}{\mathrm{TN}}_{n-1}-1& \mathrm{if}& {\mathrm{TN}}_{n-1}=\mathrm{5,4,3,2},\\ 5& \mathrm{if}& {\mathrm{TN}}_{n-1}=1\end{array}\right..$

5. in the label matching module 5 on the scene, prediction label and the actual label when the front court that receive are compared, generate mode flags signal, read last one field mode conceptual vector from memory when the front court, generate this field mode conceptual vector, and output to pattern judge module 6;

Field label matching process

According to actual label TN when the front court _n, prediction label PN when the front court _nWith last one field mode conceptual vector fldflag _N-1, available following formula is obtained the field mode conceptual vector fldflag when the front court _n, wherein, Lf represents the length of field mode conceptual vector

fldflag _n(i)＝fldflag _n-1(i+1) i＝1，2，…，Lf-1

${\mathrm{fldflag}}_n\left(\mathrm{Lf}\right)=\left\{\begin{array}{ccc}1& \mathrm{if}& T{N}_n={\mathrm{<figure-callout\; id="0"\; label="PN\; n"\; filenames="A20051004277800151.png"\; state="\{\{state\}\}">PN</figure-callout>}}_n\\ 0& \mathrm{else}& \end{array}\right..$

6. pattern judge module 6 receives and revises marking signal, a label sign and field mode conceptual vector, be 1 if revise marking signal with a label sign, and the field mode conceptual vector is omnidirectional's amount, then video is the film source video, the pattern judged result is 1, otherwise the pattern judged result is 0, the pattern judged result is outputed to the interlaced mode control system, and control goes interlaced mode to enter film mode or general mode.

The pattern evaluation algorithm

Fm _nBeing that 1 expression video is the film source video, is that 0 expression video is an ordinary video

As shown in Figure 3, revising the initial value that indicates flagMod is 1.Anomalous field number abnDif represents the number of the anomalous field that this is before continuous, and its initial value is 0.When anomalous field number abnDif=0, if abnormality mark abnorm=1, then anomalous field number abnDif adds 1, and with corrected fields diversity factor mDifC _nReplace a diversity factor difC _nWhen anomalous field number abnDif=1, if abnormality mark abnorm=1, then anomalous field number abnDif adds 1, and with corrected fields diversity factor mDifC _nReplace a diversity factor difC _n, otherwise make anomalous field number abnDif=0; When anomalous field number abnDif=2 or abnDif=3, if abnormality mark abnorm=1, then anomalous field number abnDif adds 1, otherwise makes anomalous field number abnDif=0; When anomalous field number abnDif=4, make anomalous field number abnDif=0, if abnormality mark abnorm=1, then sign flagMod=0 is revised in order, otherwise order prediction label PV=5, and to make preceding four of the field mode conceptual vector be 1.

As shown in Figure 4,,, then read the previous field data, merge into a frame if an actual label is 1,3 and 5 to being defined as the video of film source video; If an actual label is 2 and 4, then read back one field data and merge into a frame, so just 60/second interlaced video has been changed into the progressive of 60 frame/seconds.

Claims (8)

1. the detection method of 3: 2 drop-down film source videos, by comprising a difference metric module, a difference correcting module, a label generation module, a label prediction module, a label matching module and pattern judge module the video type that receives is detected, to judge whether it is 3: 2 drop-down film source videos, it is characterized in that, this method may further comprise the steps

The data input difference metric module that to work as first like parity field of front court and its front is calculated two poor in the difference metric module on the scene, the number of statistical difference dissimilarity, and export the poor correcting module of showing up as the field diversity factor;

Field difference correcting module is according to the pattern judged result of previous field, if not film mode, then the field diversity factor that receives is not handled, if film mode, then the field diversity factor is revised, field diversity factor that process is revised and preceding four corrected fields diversity factor export the label generation module of showing up, and the correction marking signal that will generate output to the pattern judge module together as current corrected fields difference vector;

Field label generation module carries out the detection of least member to the corrected fields difference vector that receives, and with the position of least member as this actual label output label matching module of showing up, if the sub-minimum of corrected fields difference vector and the ratio of least member are less than preset threshold, then label sign in field is set to 0, otherwise field label sign is set to 1, at last field label sign is outputed to the pattern judge module;

Field label prediction module reads last one actual label from memory, try to achieve the prediction label when the front court, and exports the label matching module of showing up;

In the label matching module on the scene, prediction label and the actual label when the front court that receive are compared, generate mode flags signal, read last one field mode conceptual vector from memory when the front court, generate this field mode conceptual vector, and output to the pattern judge module;

Pattern judge module received field label sign, correction marking signal and field mode conceptual vector, if a label sign and correction marking signal are 1, and the field mode conceptual vector is omnidirectional's amount, then video is the film source video, the pattern judged result is 1, otherwise the pattern judged result is 0, the pattern judged result is outputed to the interlaced mode control system, and control goes interlaced mode to enter film mode or general mode.

2. detection method as claimed in claim 1, it is characterized in that, in the poor correcting module on the scene, the method that the field diversity factor is detected is, corrected fields difference vector according to previous field is obtained the reference field diversity factor, if follow the corresponding threshold value of reference field diversity factor from the field diversity factor of field difference metric module input greater than one, think that then this diversity factor is the anomalous field diversity factor.

3. detection method as claimed in claim 1 or 2, it is characterized in that, pattern judged result according to previous field, if not film mode, then the field diversity factor that receives is not detected and revises, if film mode then detects the anomalous field diversity factor, and according to it is optionally revised.

4. detection method as claimed in claim 1 is characterized in that, the field difference of calculating two in the difference metric module on the scene is to adopt: dif _n(i, j)=| f _n(i, j)-f _N-2(i, j) |

Wherein, f _nRepresent the n field signal, f _n(i, j) expression n field signal is capable at i, the brightness value of j row, f _N-2Be n field signal f _nFirst like parity field before;

Determine the discrepancy sign according to following formula: ${\mathrm{difP}}_n(i,j)=\left\{\begin{array}{ccc}1& \mathrm{if}& {\mathrm{dif}}_n(i,j)\&GreaterEqual;\mathrm{thDif}\\ 0& \mathrm{if}& {\mathrm{dif}}_n(i,j)<\mathrm{thDif}\end{array}\right.$

The number of statistical difference dissimilarity is to adopt: ${\mathrm{difC}}_n=\underset{i,j}{\mathrm{\&Sigma;}}{\mathrm{difP}}_n(i,j)$

Wherein, discrepancy sign difP _n(i j) is 1 expression two field picture (i, j) locational some pixel value difference, discrepancy sign difP _n(i is that (i, j) a locational pixel value is identical for 0 expression two field picture j).ThDif is a fixed threshold.DifC _nAn expression diversity factor.

5. detection method as claimed in claim 1 is characterized in that, is to adopt to asking for of prediction label in the label prediction module on the scene: ${\mathrm{PN}}_n=\left\{\begin{array}{ccc}{\mathrm{TN}}_{n-1}-1& \mathrm{if}& {\mathrm{TN}}_{n-1}=\mathrm{5,4,3,2}\\ 5& \mathrm{if}& {\mathrm{TN}}_{n-1}=1\end{array}\right.$

Wherein, PN _nBe prediction label when the front court, TN _N-1It is actual label.

6. detection method as claimed in claim 1 is characterized in that, in the label matching module on the scene, the generation of current field mode marking signal is meant, ${\mathrm{fldflag}}_n\left(\mathrm{Lf}\right)=\left\{\begin{array}{ccc}1& \mathrm{if}& {\mathrm{TN}}_n={\mathrm{PN}}_n\\ 0& \mathrm{else}& \end{array}\right.$

Wherein, TN _nBe actual label when the front court, PN _nBe prediction label when the front court, fidflag _N-1Be last one field mode conceptual vector, fldflag _nBe the field mode conceptual vector when the front court, Lf represents the length of field mode conceptual vector.

7. detection method as claimed in claim 2 is characterized in that, in the poor correcting module on the scene, the evaluation of difference correction value and reference field difference is meant,

Little difference correction value: difModS=min (M _N-1)

The reference field diversity factor: $\mathrm{difModB}=\frac{(\underset{m=1}{\overset{5}{\mathrm{\&Sigma;}}}{M}_{n-1}\left(m\right)-\mathrm{difModS})}{4}$

Wherein, M _nExpression corrected fields difference vector, difModS represents little diversity factor correction value, difModB represents the reference field diversity factor.

8. detection method as claimed in claim 3 is characterized in that, determine whether need to the anomalous field difference revise adopted be, $\mathrm{mod}=\left\{\begin{array}{ccc}1& \mathrm{if}& {\mathrm{difC}}_n\&GreaterEqual;\mathrm{co}\&times;\mathrm{difModB}\\ 0& \mathrm{else}& \end{array}\right.$

Wherein, co (1＜co＜2) is a coefficient of setting in advance.

Images