CN102364933A - Motion-classification-based adaptive de-interlacing method - Google Patents

Abstract

The invention discloses a motion-classification-based adaptive de-interlacing method. Conventional interlaced scanning has the shortcomings of flickering, motion judder, vertical edge jagging and the like. In the method provided by the invention, a corresponding algorithm is adaptively selected for classification processing on the basis of performing accurate detection and granularity division on the motion state of a pixel to be interpolated, and a conventional de-interlacing algorithm is supplemented in terms of inter-field and intra-field de-interlacing by Wiener median filtering and triple median filtering. The method is more directed; and smooth transitions of de-interlacing strategies in adjacent motion states are realized, thereby achieving good de-interlacing effects.

Description

A kind of self adaptation interlace-removing method based on classification of motions

Technical field

The invention belongs to the video signal treatment technique field, relate in particular to a kind of self adaptation interlace-removing method based on classification of motions.

Background technology

Horizontal-interlace technique be exactly each two field picture be split into strange with idol, interleaved line-scanning frequency is half the when lining by line scan, the frequency spectrum of vision signal and the channel width that transmits vision signal also are line by line scan half the.Because visual persistence effect; Human eye it will be appreciated that level and smooth motion rather than the half frame images of flashing; This just under subjectivity is thought the few situation of image quality decrease, has effectively increased the utilance of channel, because the early stage communication technology is undeveloped; In order to save limited bandwidth, in conventional analogue signal TV, generally adopt interleaved technology.Interlacing scan has like this some some shortcomings: flicker, float, vertical edges are along sawtoothization etc.

Along with the development of DTV and high definition TV technology and maturation and people progressively raising to the video quality requirement; Progressive scan mode has become the preferred option of DTV scan mode, and current novel flat panel display terminal also all is the display device that support is lined by line scan.But left over a large amount of video files with the interlace mode record before this, need be converted into the video file of lining by line scan, deinterlacing technique is exactly a kind of video format conversion technology that interleaved signal is converted into progressive-scan signal.

Deinterlacing technique mainly is divided into three major types: based on the space two-dimensional interpolation technique go interlacing, based on the time 2-D interpolation technique go interlacing and based on the interlacing of going of space-time three-dimensional interpolation technology.The space two-dimensional interpolation is reduced pending pixel through the field internal information exactly; The time 2-D interpolation is exactly to utilize the contact between different to go back original image; The space-time three-dimensional interpolation is exactly to carry out interpolation with an internal information between comprehensive utilization use, and its algorithm mainly contains: content-adaptive, Motion Adaptive and motion compensation scheduling algorithm.Movement compensating algorithm can well keep the time domain and the spatial domain details of moving object, is present state-of-the-art format conversion algorithm.

And these methods all have shortcoming separately: the space two-dimensional interpolation can not increase the vertical resolution of image, and image is prone to fuzzy; The time 2-D interpolation then can be brought defectives such as sawtooth, emergence; Movement compensating algorithm is extremely responsive to motion estimation error, and interpolation makes a mistake easily.

Summary of the invention

The purpose of the inventive method is the deficiency of removing the interlacing algorithm to prior art, and a kind of self adaptation interlace-removing method based on classification of motions is provided, and avoids the interpolation mistake effectively, well goes the interlacing effect thereby obtain.

The inventive method practical implementation step is:

Step (1). at inputted video image one of them pending some C in wherein; The abscissa of judging this pending some C divided by 2 remainder whether equal this pending some C sequence number on the scene divided by 2 remainder; If both remainders equate; Constant and the execution in step (10) of gray value that then keeps this pending some C is not equal to as if both remainders, then execution in step (2).

Step (2). detect this pending some C on the scene preceding 2 with back 2 data as a reference; Calculate polarity homologous field correlation; Draw respectively this pending some C on the scene and this pending some C institute sequence number on the scene deduct 2 field correlation P and this pending some C on the scene and this pending some C sequence number on the scene add 2 correlation Q; And this pending some C preceding 1 and back 1 correlation R on the scene, calculate kinematic coefficient M

Calculating kinematic coefficient M concrete grammar is:

Calculating with the pending some C institute identical field of polarity on the scene respectively is n-2 field and the correlation P of n field and the correlation Q of n field and n+2 field, wherein pending some C on the scene be the n field.

$P=|\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i\left(n\right)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i\left(n\right)}{2}-\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i(n-2)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i(n-2)}{2}|$

$Q=|\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i\left(n\right)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i\left(n\right)}{2}-\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i(n+2)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i(n+2)}{2}|$

Wherein, T _i(n) be last 1 row, 1 * 3 block of pixels with the pending some C in n field, by T _-1 (n), T ₀(n), T ₁(n) 3 compositions;

B _i(n) be following 1 row, 1 * 3 block of pixels with the pending some C in n field, by B _-1(n), B ₀(n), B ₁(n) 3 compositions;

Calculate and be inserted into some an institute opposite polarity n-1 field on the scene and a n+1 field correlation R simultaneously

$R=|\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{X}_i(n-1)-\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{X}_i(n+1)|$

X _-1(n), X ₀(n), X ₁(n) be same delegation composition 1 * 3 block of pixels of the pending some C in n field;

Kinematic coefficient M can get thus

Step (3). with the kinematic coefficient M of step (2) gained and preset movement threshold 1 relatively,, then the confidence level of low-speed motion is made as 1 and execution in step (4) if gained kinematic coefficient M is less than preset movement threshold 1; Otherwise, the confidence level of low-speed motion is made as 0 and execution in step (6);

Step (4). adopt the method for block of pixels match search; Form 1 * 3 search block at 2 about will comprising that pending some C go together with it; With pending some C this position of previous field on the scene search block and pending some C back 3 search block in last 2 row left, center, rights, institute on the scene be expert at about 2 search block and down 3 search block in left, center, right of 2 row carry out absolute and SAD calculating respectively; Result of calculation and movement threshold 0 are compared: if result of calculation is greater than movement threshold 0; The confidence level of then slightly moving is set to 1; And deduct the motion vector that previous field search block vector is made as pending some C, and execution in step (5) with a back best match search piece vector; Otherwise slightly the confidence level of motion is set to 0, and execution in step (8), and a wherein back best match search piece is the pairing search block of result of calculation;

Step (5). adopt dimension to receive median filtering method pending some C of small size motion carried out interpolation, and execution in step (10);

The concrete steps that pending some C carries out interpolation are:

A. get A on the direction of motion of pending some C place, n field, 3 points of B and A ', 3 points wherein being got are respectively preceding and back 1 point on pending the C direction of motion at 2;

B. A point, B point and pending some C being carried out Wiener filtering calculates:

F _wiener＝wiener(A，B，C)

Wherein $\left[\begin{array}{ccc}{r}_x\left(0\right)& r_x^{*}\left(1\right)& r_x^{*}\left(2\right)\\ r_x\left(1\right)& r_x\left(0\right)& r_x^{*}\left(1\right)\\ r_x\left(2\right)& r_x\left(1\right)& r_x\left(0\right)\end{array}\right]\left[\begin{array}{c}w\left(0\right)\\ w\left(1\right)\\ w\left(2\right)\end{array}\right]=\left[\begin{array}{c}{r}_x\left(1\right)\\ r_x\left(2\right)\\ r_x\left(3\right)\end{array}\right]$

F _wiener＝w(0)*C+w(1)*B+w(2)*A

Wherein, F _WienerCalculate gained result, r for carrying out Wiener filtering _x, r _x ^*Be respectively the auto-correlation coefficient and the conjugation auto-correlation coefficient of sequence [A, B, C, A '], w (0), w (1), w (2) is respectively the weight coefficient of A point, B point and pending some C.

C. get adjacent 2 D and E in pending the vertical direction of the C direction of motion, calculate corresponding interpolation result:

S _low(i，j)＝median(F _wiener，E，D)

S wherein _Low(i j) is F _Wiener, E and D median;

Step (6). the kinematic coefficient M and the preset movement threshold 2 of step (2) gained are compared: if the kinematic coefficient M of gained is greater than preset movement threshold 2; The confidence level that then will significantly move is made as 1 and execution in step (7), otherwise the confidence level that will significantly move is made as 0 and execution in step (9);

Step (7). pending some C to affiliated uses triple medium filterings to carry out interpolation, then execution in step (10);

The concrete steps that pending some C carries out triple medium filtering interpolation are:

D establish pending some C coordinate (j, i), the block of pixels with 3 * 3 is carried out medium filtering to three capable points of j-1:

F ₁＝median(S(i-1，j-1)，S(i，j-1)，S(i+1，j-1))

F ₁For S (i-1, j-1), S (i, j-1) and S (i+1, median j-1);

Three points capable to j+1 carry out medium filtering:

F ₂＝median(S(i-1，j+1)，S(i，j+1)，S(i+1，j+1))

F ₂For S (i-1, j+1), S (i, j+1) and S (i+1, median j+1);

E to S (i-1, j), F ₁, F ₂Carry out medium filtering, obtain the interpolation result S of the pending some C in non-edge under significantly moving _High(i, j):

S _high(i，j)＝median(S(i-1，j)，F ₁，F ₂)；

Step (8). interpolation is carried out in affiliated pending some C use occasion and method, then and execution in step (10);

Step (9). for affiliated pending some C use respectively step (7) and step (8) must show up interior and between the interpolation result, calculate the space-time weight again, merge two kinds of interpolation results with the space-time weight;

Space-time weights α is:

$\mathrm{\α}=[\cos (\mathrm{\π}\·\frac{M-T_1}{T_2-T_1})+1]/2$

T wherein ₁ Be movement threshold 1, T ₂Be movement threshold 2,

Y _intp＝α·Y _inter+(1-α)·Y _intra

Wherein, Y _InterBe the interpolation result that utilizes algorithm between the field, Y _IntraBe to utilize an interpolation result of interior algorithm, Y _IntpBe the interpolation result that finally calculates;

Step (10). the said inputted video image of traversal step (1) judges whether also to exist pending point: if exist, then return step (1) the pending point of the next one is handled; If do not exist, then finish.

Compared with prior art, beneficial effect of the present invention:

(1) the present invention is through accurately detecting in the motion state of treating interpolating pixel point on the basis of dividing with fine granularity; Select respective algorithms to carry out classification processing adaptively; Interpolation has more specific aim, and the interlacing strategy that goes under the adjacent motion state is seamlessly transitted.

(2) the present invention not only improves traditional Motion Adaptive algorithm; And dimension is wherein received medium filtering and triple medium filtering respectively from going the interlacing algorithm to replenish with in-field deinterlacing two aspects to tradition between the field; Can avoid the interpolation mistake effectively, well go the interlacing effect thereby obtain.

Description of drawings

Fig. 1 is the self adaptation interlace-removing method flow chart based on classification of motions;

Fig. 2 is five motion detection sketch mapes;

Fig. 3 is the division sketch map of low-speed motion state;

Fig. 4 is the detection sketch map of the direction of motion.

Fig. 5 receives the medium filtering sketch map for dimension;

Fig. 6 is triple medium filtering sketch mapes.

Embodiment

Below in conjunction with accompanying drawing the present invention is described further

As shown in Figure 1, the inventive method practical implementation step is:

Step (1). at inputted video image one of them pending some C in wherein; The abscissa of judging this pending some C divided by 2 remainder whether equal this pending some C sequence number on the scene divided by 2 remainder; If both remainders equate; Constant and the execution in step (10) of gray value that then keeps this pending some C is not equal to as if both remainders, then execution in step (2).

Step (2). detect this pending some C on the scene preceding 2 with back 2 data as a reference; Calculate polarity homologous field correlation; Draw respectively this pending some C on the scene and this pending some C institute sequence number on the scene deduct 2 field correlation P and this pending some C on the scene and this pending some C sequence number on the scene add 2 correlation Q; And this pending some C preceding 1 and back 1 correlation R on the scene, calculate kinematic coefficient M

As shown in Figure 2, calculate kinematic coefficient M concrete grammar and be:

Calculating with the pending some C institute identical field of polarity on the scene respectively is n-2 field and the correlation P of n field and the correlation Q of n field and n+2 field, wherein pending some C on the scene be the n field.

$P=|\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i\left(n\right)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i\left(n\right)}{2}-\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i(n-2)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i(n-2)}{2}|$

$Q=|\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i\left(n\right)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i\left(n\right)}{2}-\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i(n+2)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i(n+2)}{2}|$

Wherein, T _i(n) be last 1 row, 1 * 3 block of pixels with the pending some C in n field, by T _-1(n), T ₀(n), T ₁(n) 3 compositions;

B _i(n) be following 1 row, 1 * 3 block of pixels with the pending some C in n field, by B _-1(n), B ₀(n), B ₁(n) 3 compositions;

Calculate and be inserted into some an institute opposite polarity n-1 field on the scene and a n+1 field correlation R simultaneously

$R=|\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{X}_i(n-1)-\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{X}_i(n+1)|$

X _-1(n), X ₀(n), X ₁(n) be same delegation composition 1 * 3 block of pixels of the pending some C in n field;

Kinematic coefficient M can get thus

Step (3). with the kinematic coefficient M of step (2) gained and preset movement threshold 1 relatively,, then the confidence level of low-speed motion is made as 1 and execution in step (4) if gained kinematic coefficient M is less than preset movement threshold 1; Otherwise, the confidence level of low-speed motion is made as 0 and execution in step (6);

Step (4). as shown in Figure 3; Adopt the method for block of pixels match search; Form 1 * 3 search block at 2 about will comprising that pending some C go together with it; With pending some C this position of previous field on the scene search block and pending some C back 3 search block in last 2 row left, center, rights, institute on the scene be expert at about 2 search block and down 3 search block in left, center, right of 2 row carry out absolute and SAD calculating respectively; Result of calculation and movement threshold 0 are compared: if result of calculation is greater than movement threshold 0; The confidence level of then slightly moving is set to 1, and deducts the motion vector that previous field search block vector is made as pending some C with a back best match search piece vector, and execution in step (5); Otherwise slightly the confidence level of motion is set to 0, and execution in step (8), and a wherein back best match search piece is the pairing search block of result of calculation;

Step (5). adopt dimension to receive median filtering method pending some C of small size motion carried out interpolation, and execution in step (10);

Like Fig. 4 and shown in Figure 5, the concrete steps that pending some C carries out interpolation are:

Get A on the direction of motion of pending some C place, a n field, 3 points of B and A ', 3 points wherein being got are respectively preceding and back 1 point on pending the C direction of motion at 2;

B carries out Wiener filtering to A point, B point and pending some C and calculates:

F _wiener＝wiener(A，B，C)

Wherein $\left[\begin{array}{ccc}{r}_x\left(0\right)& r_x^{*}\left(1\right)& r_x^{*}\left(2\right)\\ r_x\left(1\right)& r_x\left(0\right)& r_x^{*}\left(1\right)\\ r_x\left(2\right)& r_x\left(1\right)& r_x\left(0\right)\end{array}\right]\left[\begin{array}{c}w\left(0\right)\\ w\left(1\right)\\ w\left(2\right)\end{array}\right]=\left[\begin{array}{c}{r}_x\left(1\right)\\ r_x\left(2\right)\\ r_x\left(3\right)\end{array}\right]$

F _wiener＝w(0)*C+w(1)*B+w(2)*A

Wherein, F _WienerCalculate gained result, r for carrying out Wiener filtering _x, r _x ^*Be respectively the auto-correlation coefficient and the conjugation auto-correlation coefficient of sequence [A, B, C, A '], w (0), w (1), w (2) is respectively the weight coefficient of A point, B point and pending some C.

C gets adjacent 2 D and E in pending the vertical direction of the C direction of motion, calculates corresponding interpolation result:

S _low(i，j)＝median(F _wiener，E，D)

S wherein _Low(i j) is F _Wiener, E and D median;

Step (6). the kinematic coefficient M and the preset movement threshold 2 of step (2) gained are compared: if the kinematic coefficient M of gained is greater than preset movement threshold 2; The confidence level that then will significantly move is made as 1 and execution in step (7), otherwise the confidence level that will significantly move is made as 0 and execution in step (9);

Step (7). pending some C to affiliated uses triple medium filterings to carry out interpolation, then execution in step (10);

As shown in Figure 6, the concrete steps that pending some C carries out triple medium filtering interpolation are:

D establish pending some C coordinate (j, i), the block of pixels with 3 * 3 is carried out medium filtering to three capable points of j-1:

F ₁＝median(S(i-1，j-1)，S(i，j-1)，S(i+1，j-1))

F ₁For S (i-1, j-1), S (i, j-1) and S (i+1, median j-1);

Three points capable to j+1 carry out medium filtering:

F ₂＝median(S(i-1，j+1)，S(i，j+1)，S(i+1，j+1))

F ₂For S (i-1, j+1), S (i, j+1) and S (i+1, median j+1);

E to S (i-1, j), F ₁, F ₂Carry out medium filtering, obtain the interpolation result S of the pending some C in non-edge under significantly moving _High(i, j):

S _high(i，j)＝median(S(i-1，j)，F ₁，F ₂)；

Step (8). interpolation is carried out in affiliated pending some C use occasion and method, then and execution in step (10);

Step (9). for affiliated pending some C use respectively step (7) and step (8) must show up interior and between the interpolation result, calculate the space-time weight again, merge two kinds of interpolation results with the space-time weight;

Space-time weights α is:

$\mathrm{\α}=[\cos (\mathrm{\π}\·\frac{M-T_1}{T_2-T_1})+1]/2$

T wherein ₁Be movement threshold 1, T ₂Be movement threshold 2,

Y _intp＝α·Y _inter+(1-α)·Y _intra

Wherein, Y _InterBe the interpolation result that utilizes algorithm between the field, Y _IntraBe to utilize an interpolation result of interior algorithm, Y _IntpBe the interpolation result that finally calculates;

Step (10). the said inputted video image of traversal step (1) judges whether also to exist pending point: if exist, then return step (1) the pending point of the next one is handled; If do not exist, then finish.

Claims (1)

1. the self adaptation interlace-removing method based on classification of motions is characterized in that this method comprises the steps:

Step (1). at inputted video image one of them pending some C in wherein; The abscissa of judging this pending some C divided by 2 remainder whether equal this pending some C sequence number on the scene divided by 2 remainder; If both remainders equate; Constant and the execution in step (10) of gray value that then keeps this pending some C is not equal to as if both remainders, then execution in step (2);

Step (2). detect this pending some C on the scene preceding 2 with back 2 data as a reference; Calculate polarity homologous field correlation; Draw respectively this pending some C on the scene and this pending some C institute sequence number on the scene deduct 2 field correlation P and this pending some C on the scene and this pending some C sequence number on the scene add 2 correlation Q; And this pending some C preceding 1 and back 1 correlation R on the scene, calculate kinematic coefficient M

Calculating kinematic coefficient M concrete grammar is:

Calculating with the pending some C institute identical field of polarity on the scene respectively is n-2 field and the correlation P of n field and the correlation Q of n field and n+2 field, wherein pending some C on the scene be the n field;

$P=|\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i\left(n\right)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i\left(n\right)}{2}-\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i(n-2)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i(n-2)}{2}|$

$Q=|\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i\left(n\right)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i\left(n\right)}{2}-\frac{\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{T}_i(n+2)+\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{B}_i(n+2)}{2}|$

Wherein, T _i(n) be last 1 row, 1 * 3 block of pixels with the pending some C in n field, by T _-1(n), T ₀(n), T ₁(n) 3 compositions;

B _i(n) be following 1 row, 1 * 3 block of pixels with the pending some C in n field, by B _-1(n), B ₀(n), B ₁(n) 3 compositions;

Calculate and be inserted into some an institute opposite polarity n-1 field on the scene and a n+1 field correlation R simultaneously

$R=|\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{X}_i(n-1)-\underset{i=-1}{\overset{1}{\mathrm{\Σ}}}{X}_i(n+1)|$

X _-1(n), X ₀(n) X ₁(n) be same delegation composition 1 * 3 block of pixels of the pending some C in n field; Kinematic coefficient M can get thus

Step (3). with the kinematic coefficient M of step (2) gained and preset movement threshold 1 relatively,, then the confidence level of low-speed motion is made as 1 and execution in step (4) if gained kinematic coefficient M is less than preset movement threshold 1; Otherwise, the confidence level of low-speed motion is made as 0 and execution in step (6);

Step (4). adopt the method for block of pixels match search; Form 1 * 3 search block at 2 about will comprising that pending some C go together with it; With pending some C this position of previous field on the scene search block and pending some C back 3 search block in last 2 row left, center, rights, institute on the scene be expert at about 2 search block and down 3 search block in left, center, right of 2 row carry out absolute and SAD calculating respectively; Result of calculation and movement threshold 0 are compared: if result of calculation is greater than movement threshold 0; The confidence level of then slightly moving is set to 1; And deduct the motion vector that previous field search block vector is made as pending some C, and execution in step (5) with a back best match search piece vector; Otherwise slightly the confidence level of motion is set to 0, and execution in step (8), and a wherein back best match search piece is the pairing search block of result of calculation;

Step (5). adopt dimension to receive median filtering method pending some C of small size motion carried out interpolation, and execution in step (10);

The concrete steps that pending some C carries out interpolation are:

A. get A on the direction of motion of pending some C place, n field, 3 points of B and A ', 3 points wherein being got are respectively preceding and back 1 point on pending the C direction of motion at 2;

B. A point, B point and pending some C being carried out Wiener filtering calculates:

F _wiener＝wiener(A，B，C)

Wherein $\left[\begin{array}{ccc}{r}_x\left(0\right)& r_x^{*}\left(1\right)& r_x^{*}\left(2\right)\\ r_x\left(1\right)& r_x\left(0\right)& r_x^{*}\left(1\right)\\ r_x\left(2\right)& r_x\left(1\right)& r_x\left(0\right)\end{array}\right]\left[\begin{array}{c}w\left(0\right)\\ w\left(1\right)\\ w\left(2\right)\end{array}\right]=\left[\begin{array}{c}{r}_x\left(1\right)\\ r_x\left(2\right)\\ r_x\left(3\right)\end{array}\right]$

F _wiener＝w(0)*C+w(1)*B+w(2)*A

Wherein, F _WienerCalculate gained result, r for carrying out Wiener filtering _x, r _x ^*Be respectively the auto-correlation coefficient and the conjugation auto-correlation coefficient of sequence [A, B, C, A '], w (0), w (1), w (2) is respectively the weight coefficient of A point, B point and pending some C;

C. get adjacent 2 D and E in pending the vertical direction of the C direction of motion, calculate corresponding interpolation result:

S _low(i，j)＝median(F _wiener，E，D)

S wherein _Low(i j) is F _Wiener, E and D median;

Step (6). the kinematic coefficient M and the preset movement threshold 2 of step (2) gained are compared: if the kinematic coefficient M of gained is greater than preset movement threshold 2; The confidence level that then will significantly move is made as 1 and execution in step (7), otherwise the confidence level that will significantly move is made as 0 and execution in step (9);

Step (7). pending some C to affiliated uses triple medium filterings to carry out interpolation, then execution in step (10);

The concrete steps that pending some C carries out triple medium filtering interpolation are:

D establish pending some C coordinate (j, i), the block of pixels with 3 * 3 is carried out medium filtering to three capable points of j-1:

F ₁＝median(S(i-1，j-1)，S(i，j-1)，S(i+1，j-1))

F ₁For S (i-1, j-1), S (i, j-1) and S (i+1, median j-1);

Three points capable to j+1 carry out medium filtering:

F ₂＝median(S(i-1，j+1)，S(i，j+1)，S(i+1，j+1))

F ₂For S (i-1, j+1), S (i, j+1) and S (i+1, median j+1);

E to S (i-1, j), F ₁, F ₂Carry out medium filtering, obtain the interpolation result S of the pending some C in non-edge under significantly moving _High(i, j):

S _high(i，j)＝median(S(i-1，j)，F ₁，F ₂)；

Step (8). interpolation is carried out in affiliated pending some C use occasion and method, then and execution in step (10);

Step (9). for affiliated pending some C use respectively step (7) and step (8) must show up interior and between the interpolation result, calculate the space-time weight again, merge two kinds of interpolation results with the space-time weight;

Space-time weights α is:

$\mathrm{\α}=[\cos (\mathrm{\π}\·\frac{M-T_1}{T_2-T_1})+1]/2$

T wherein ₁Be movement threshold 1, T ₂Be movement threshold 2,

Y _intp＝α·Y _inter+(1-α)·Y _intra

Wherein, Y _InterBe the interpolation result that utilizes algorithm between the field, Y _IntraBe to utilize an interpolation result of interior algorithm, Y _IntpBe the interpolation result that finally calculates;

Step (10). the said inputted video image of traversal step (1) judges whether also to exist pending point: if exist, then return step (1) the pending point of the next one is handled; If do not exist, then finish.

Images