CN101344968A - Movement compensation method for star sky background image - Google Patents

Abstract

The invention discloses a motion compensation method of a starry sky background image. The method comprises the steps that: first, according to the gray scale and the motion feature of the starry sky background image, motion background compensation is carried out by utilizing two key processes of sample motion parameter estimation and integral motion parameter estimation. An abnormal point is directly removed in the sample motion vector estimation stage and the amount of calculating redundant information in the motion compensation method is reduced; and second, the method of the combination of block matching and centroid extraction is introduced in the sample motion parameter estimation stage. Due to the adoption of the block matching method, the operation speed of sample estimation is raised, the removal of the abnormal point furthermore reduces the amount of data information to be processed and the method based on the centroid extraction further improves the precision of a sample point estimation parameter. The motion compensation method adopts the method based on the combination of the block matching and the centroid extraction to achieve motion compensation in the starry sky background, therefore, calculation amount is lowered by 50 percent and the compensation precision is improved to 0.5 pixels from 1 pixel of the prior art.

Description

Motion compensation process in the starry sky background image

Technical field

The present invention relates to the motion compensation process in a kind of the starry sky background image disposal route, particularly starry sky background image.

Background technology

Weak target under the compound movement background detects problem and has directly determined the operating distance of astronomical sight system and detected performance, and the movement background compensation is one of gordian technique of Weak target detection.

Document " based on the movement background compensation technique of robust regression; signal Processing; 2002; 18 (1): 36-38 " discloses a kind of estimation of carrying out the sample point kinematic parameter based on the method for light stream, and employing iterates the estimation that weighted least-squares method is carried out the global motion parameter, the error of the global motion parameter estimation of having avoided impact point to be taken as sampled point and having introduced, but wherein asking for of weight function increased operand greatly, reduce the speed of motion compensation, also can only realize the compensation precision of 1 pixel.

Summary of the invention

For the deficiency that speed is low, precision is low that overcomes the prior art motion compensation, the invention provides the motion compensation process in a kind of starry sky background image, employing is based on the method for piece coupling, can effectively improve the arithmetic speed that sample is estimated, the rejecting of abnormity point can reduce the quantity of information of pending data, can improve the precision of sample point estimated parameter based on the method for barycenter extraction.

The technical solution adopted for the present invention to solve the technical problems: the motion compensation process in a kind of starry sky background image is characterized in that comprising the steps:

(a) determining by effective coverage between consecutive frame, image is evenly divided, carry out choosing of sample point in the little module after division, and mate the mode that combines with centroid method by piece, carry out the estimation of sample point motion vector, remove redundant information by abnormity point elimination, calculate the motion vector of sample characteristics;

(b) employing determines to characterize the camera motion of global motion model based on six parameter models of parallel projection;

(c) the definition sample point is gathered, and calculates the kinematic parameter of all sample points, goes out velocity field by the camera motion model parameter estimation that makes up, and obtains the kinematic parameter of global motion model by the optimum solution of finding the solution global motion model parameters;

(d). compensate by the back two field picture of bilinear interpolation method adjacent two frames.

The invention has the beneficial effects as follows: owing to adopt the method for mating based on piece, improved the arithmetic speed that sample is estimated effectively, the rejecting of abnormity point has further reduced the quantity of information of pending data again, and further improved the precision of sample point estimated parameter based on the method that barycenter extracts, the method that the present invention's employing is mated based on piece and the barycenter extraction combines is waken up with a start the motion compensation under the starry sky background, operand has reduced by 50%, and compensation precision has been brought up to 0.5 pixel by 1 original pixel.

Below in conjunction with drawings and Examples the present invention is elaborated.

Description of drawings

Accompanying drawing is the motion compensation process process flow diagram in the starry sky background image of the present invention.

Embodiment

With reference to accompanying drawing.1, at first carries out the sample characteristics estimation.

The deep space background shows as black in image, fixed star is similar to small and weak moving target, is rendered as point source in image, because also there is much noise in detecting devices self in the image.Take into full account the characteristic of image and starry sky background motion target detection, directly carry out the rejecting of abnormity point in the Sample selection stage, avoid the employing of camera model motion model parameters estimation stages to iterate weighted least-squares method and carried out asking for of global parameter, saved finding the solution of weight function, simplified the process that global parameter is asked for, effectively shorten the time of background compensation, reduced calculated amount.

The basic process in sample characteristics estimation stage is as follows:

(a) effective coverage is determined between consecutive frame.

Because background generation background motion in the video sequence is promptly worked as in the boundary effect influence, there is relative motion in background between consecutive frame, and image can not overlap fully.Therefore, have only in sample point is chosen at the public domain, could effectively estimate the kinematic parameter of sample point.Therefore, at first need to determine effective coverage between adjacent two frames.

The scope of choosing the border horizontal ordinate of effective coverage is 16≤x≤L _x-16, the scope of ordinate is 16≤y≤L _y-16, wherein 16 expressions are decided to be 16 pixels with border width, wherein, and L _xAnd L _yDifference presentation video frame size.Therefore, can obtain the effective coverage of consecutive frame.

(b) based on uniform sampling thought, image evenly is divided into 9, in little module, carry out sample point and choose.

More representative and dispersed for sample point is in earlier stage chosen, it is more accurate to make global motion model parameters estimate, avoids choosing too much sample point and consumes the more processing time, and the present invention has used for reference the thought of uniform sampling.

The image of present frame evenly is divided into 9, uses for reference the notion of magnitude in the astronomy, the global search gray-scale value is greater than the sampled point of given threshold value, and this moment, threshold value was decided to be 180.With evenly divide threshold value in the fritter of back greater than 180 some as the interested feature point set that finds, record searches the gray-scale value of sampled point and horizontal stroke, the ordinate of position at first, with this as the sampled point of selecting.

(c) characteristic matching and estimation of motion vectors.

The present invention has adopted the technology that combines with centroid method based on the piece coupling to carry out the estimation of motion vectors of sample point.Promptly adopt the matching criterior of minimum average B configuration absolute difference function, utilize three step search procedures to carry out the optimum matching block search, wherein, the size of determining piece is 16 * 16.Simultaneously, in order to reduce the operation times of each position to be matched in the piece matching process, the precision of maintainance block coupling has been introduced continuous eliminating algorithm, and the precision of registration has been brought up to 0.5 pixel by 1 pixel.

(d) abnormity point elimination.

Under the starry sky background, moving target and background fixed star all show as hot spot in image, but target has different kinetic characteristics with background, if moving target is chosen as sample point, to estimating at than mistake of global motion parameter.Owing to deep space background under the starry sky background shows as black region in image,, possibly can't find best matching blocks again if this point is chosen as sample point when carrying out search matched.We claim these points to be abnormity point.

In order effectively to avoid abnormity point to bring than mistake, will reject abnormity point in this stage to overall motion estimation.8bit image with 1024 * 1024 be example.At first, seek gray-scale value, gray scale is rejected as abnormity point less than 180 point greater than 180 point; Secondly, the motion vector that estimates is judged, changed in, then carry out further work less than 3 pixel coverages, otherwise, be made as abnormity point and reject, and in this module, seek unique point again and carry out estimation of motion vectors.Employing is with the rejecting of carrying out abnormity point with the mode of motion vector comparison of measuring of sampled point and reference point, avoided effectively iterating the method that weighted least-squares method is carried out abnormity point elimination as the use of being mentioned in the background technology, reduced by 50% calculated amount, improved the efficient of calculating.

(e) the accurate estimation of sample characteristics kinematic parameter.

By abnormity point elimination, selected sample point is the point on the fixed star.Utilize the pixel threshold method that mark is carried out in the zone at the 1st frame sample point place earlier, determine the border of fixed star spot, utilize the square weighting method again,, carry out barycenter and ask for as formula (1), wherein, x, y are the horizontal ordinate of pixel, F (x, y) be that (m, n are the size of window for x, the gray-scale value of y) locating.x ₀, y ₀Be the center-of-mass coordinate of trying to achieve.

$x_0=\frac{\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}{F}^2(x,y)x}{\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}{F}^2(x,y)}$ $y_0=\frac{\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}{F}^2(x,y)y}{\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}{F}^2(x,y)}---\left(1\right)$

Then the motion vector that draws with Block Matching Algorithm (u v) draws the position of this sample point in the 2nd frame, promptly (x+u, y+v), and calculate the motion after fixed star centroid position (x ' ₀, y ' ₀).Therefore, $(\hat{u},\hat{v})=(x_0^{\′}-x_0,y_0^{\′}-y_0)$ Be the sample characteristics accurate motion vectors that estimates.

2, camera motion model.

The global motion model parameters of computed image needs definite camera motion model that characterizes global motion, considers the validity of complexity of calculation and algorithm, and the present invention adopts six parameter models based on parallel projection.

When mode of motion such as image existence rotation, translation, system can adopt the motion model parameters estimation technique, and motion is described to Three-dimension Target.Typical kinematic parameter estimation model as shown in Equation (2).

Suppose at moment t _kAny coordinate of certain of imaging rigid body be (X, Y, Z), at moment t _K+1This point moves to (X ', Y ', Z '), and the translation that rotation that the available matrix R of this motion describes and vector T represent is synthesized, and has:

$\left[\begin{array}{c}{X}^{\′}\\ Y^{\′}\\ Z^{\′}\end{array}\right]=R\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]+T=\left[\begin{array}{c}{r}_{11}{r}_{12}{r}_{13}\\ r_{21}{r}_{22}{r}_{23}\\ r_{31}{r}_{32}{r}_{33}\end{array}\right]\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]+\left[\begin{array}{c}{t}_1\\ t_2\\ t_3\end{array}\right]---\left(2\right)$

Impact point coordinate in imaging plane (x, y) with target three-dimensional point coordinate (X, Y, relation Z) such as formula (3):

(x，y)＝(X，Y)(x′，y′)＝(X′，Y′) (3)

Can obtain formula (4) by formula (2) and formula (3)

$\left\{\begin{array}{c}{x}^{\′}=r_{11}x+r_{12}y+(r_{13}Z+t_1)\\ y^{\′}=r_{21}x+r_{22}y+(r_{23}Z+t_2)\end{array}\right.---\left(4\right)$

Further can be expressed as formula (5):

$\left\{\begin{array}{c}{x}^{\′}=a_{1}x+a_{2}y+a_3\\ y^{\′}=a_{4}x+a_{5}y+a_6\end{array}\right.---\left(5\right)$

Wherein, a ₁=r ₁₁, a ₂=r ₁₂, a ₄=r ₂₁, a ₅=r ₂₂, a ₃=r ₁₃Z+t ₁, a ₆=r ₂₃Z+t ₂

Therefore, only need seek out optimum [a ₁, a ₂, a ₃, a ₄, a ₅, a ₆] promptly obtained global motion model parameters.

3, global motion model parameters is asked for.

As shown in Equation (6), definition S represents the set of sample point, $\hat{V}=\{{\hat{V}}_1,L,{\hat{V}}_s,L,{\hat{V}}_n|s\∈S\}$ The set of all sample point kinematic parameters that expression is calculated,

Expression sampled point s place is by the estimated velocity field that goes out of the camera motion model of formula (5) description, as shown in Equation (7).

$\hat{V}=\{{\hat{V}}_1,L,{\hat{V}}_s,L,{\hat{V}}_n|s\∈S\}---\left(6\right)$

$\hat{U}(x,A)=\left[\begin{array}{c}{\hat{u}}_x(s,A)\\ {\hat{u}}_y(s,A)\end{array}\right]=\left[\begin{array}{c}{a}_1{x}_s+a_2{y}_s+a_3\\ a_4{x}_s+a_5{y}_s+a_6\end{array}\right]---\left(7\right)$

Optimum model parameter C is found the solution in being solved to of global motion model parameters _Opt, i.e. the error of minimum model.As formula (8).

$C_{\mathrm{opt}}=\underset{A\∈R^6}{\mathrm{Arg}\min }\underset{s\∈S}{\mathrm{\Σ}}\mathrm{\ρ}({\hat{V}}_s-\hat{U}(s,A))---\left(8\right)$

Wherein, ρ (g) is the optimal estimation function, adopts least square method.

4, movement background compensation.

Parameter [a when camera model ₁, a ₂, a ₃, a ₄, a ₅, a ₆] be after global motion model parameters estimates, selecting with the former frame is benchmark, and the background of back one frame is carried out motion compensation.

Adopt the bilinear interpolation method method.Wherein, I (x+u, y+v, k+1) choose near (x+u, y+v) brightness value of four of the position pixels makes up, the image after the motion compensation obtains by formula (9).

I(x+u，y+v，k+1)＝(1-α _x)(1-α _y)I(x ₀，y ₀，k+1)+α _x(1-α _y)I(x ₀+1，y ₀，k+1) (9)

+(1-α _x)α _yI(x ₀，y ₀+1，k+1)+α _xα _yI(x ₀+1，y ₀+1，k+1)

Wherein, x ₀Be the integral part of x+u, α _xFraction part for x+u; y ₀Be the integral part of y+v, α _yFraction part for y+v.

Claims (1)

1, the motion compensation process in a kind of starry sky background image is characterized in that comprising the steps:

(a) determining by effective coverage between consecutive frame, image is evenly divided, carry out choosing of sample point in the little module after division, and mate the mode that combines with centroid method by piece, carry out the estimation of sample point motion vector, remove redundant information by abnormity point elimination, calculate the motion vector of sample characteristics;

(b) employing determines to characterize the camera motion of global motion model based on six parameter models of parallel projection;

(c) the definition sample point is gathered, and calculates the kinematic parameter of all sample points, goes out velocity field by the camera motion model parameter estimation that makes up, and obtains the kinematic parameter of global motion model by the optimum solution of finding the solution global motion model parameters;

(d) compensate by the back two field picture of bilinear interpolation method adjacent two frames.

Images