CN100545640C - A kind of automatic tracking method for video frequency microscopic image cell - Google Patents

Abstract

The present invention relates to a kind of method of video microscopic image cell automatic tracking.Existing tracking cell method automaticity is lower, can not adapt to the property complicated and changeable and the many cells motion tracking of cell movement form.Step of the present invention comprises: the cell movement video frequency microscopic image that obtains is carried out enhancement process frame by frame; Cell movement extracting target from images cell after the enhancement process; Set up the cell movement kinetic model; The target cell of pursuit movement.On the basis of the present invention noise and interference characteristic in analysis of cells kinetic characteristic and image, adopt dynamic system modeling and stochastic modeling method, by recursion Bayesian filtering and data association technology, the movement locus of pair cell is followed the tracks of, and automaticity height, processing power are strong.

Description

A kind of automatic tracking method for video frequency microscopic image cell

Technical field

The invention belongs to cell biology and field of biological pharmacy, relate to a kind of method of video microscopic image cell automatic tracking.

Background technology

Tracking cell is meant that in observation time attributes such as movement locus to specific cells, movement velocity, color, shape carry out qualitative or quantitative analysis, it be carry out cytoactive, cell is migrated and the effective ways and the indispensable means of cell biology such as cytotaxis and bio-pharmaceuticals research, all has crucial Research Significance and practical value aspect biology, pharmacology and the pathology.

The implementation method of present domestic tracking cell mainly still is artificial timing detection, record under microscopy apparatus is assisted, and its automaticity is lower.Require the staff to pay arduous work on the one hand, have higher labour intensity; On the other hand since factors such as the noise of the property complicated and changeable (as cell division, combination, gathering, intersection) of cell movement form and video image itself and interference make the artificial observation process difficult more, and higher error rate arranged.

In recent years, foreign study mechanism has obtained some achievements in research aspect cell automatic tracking, develop a collection of cell automatic tracking system (as University of Virginia, University of Aberdeen and European Molecular Biology Laboratory etc.) by image analysis software, but that is that all right aspect commercialization is ripe.Though these equipment can calculate the position of cell on image by Flame Image Process and analysis software, for the division of cell be difficult to handle in conjunction with complicated phenomenons such as intersecting of the separating of, cell aggregation and cell mass, many cells movement locus.

Summary of the invention

The method that provides a kind of automaticity height, video frequency microscopic image cell that processing power is strong to follow the tracks of for area researches such as cell biology and bio-pharmaceuticals is provided at the deficiency of existing cell with method.

The present invention includes following steps: 1. obtain real-time cell movement image by phase microscope, the cell movement video frequency microscopic image that obtains is carried out enhancement process frame by frame; 2. the cell movement extracting target from images cell after the enhancement process; 3. set up the cell movement kinetic model; 4, the target cell of pursuit movement.

The enhancement process of video frequency microscopic image adopts generalized fuzzy enhancement process method in the step 1, and concrete steps are as follows:

1) pending image f is imported in initialization, and the initial value of setting iterations is r=1;

2) adopt median filter to carry out smothing filtering, (i j) as the center of window (can cover the unit of certain image), covers visual pixel grey scale mean value as pixel f (i, new gray-scale value f j) with window with each pixel f of image f _Ij

3) with new gray-scale value f _IjCarry out iteration the r time, determine the fuzzy characteristics plane { μ of image behind the medium filtering _Ij(r) };

4) to μ _Ij(r) make following nonlinear transformation, transformation results is designated as μ ' _Ij(r);

${\mathrm{\&mu;}}_{\mathrm{ij}}^{\&prime;}\left(r\right)=T\left({\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right)\right)=\left\{\begin{array}{cc}2{\left({\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right)\right)}^2& 0\&le;{\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right)\&le;0.5\\ 1-2{(1-{\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right))}^2& 0.5<{\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right)\&le;1\end{array}\right.$

5) to μ ' _Ij(r) do inverse transformation, obtain new gray level image f ' _Ij;

6) suitable f is set in greyscale transformation _Min ^e, f _Max ^e, calculate the gradation of image { f after generalized fuzzy strengthens _Ij ^e;

7) relatively strengthen the image quality in images evaluation index the r time and the r-1 time, if σ _w(r) less than σ _w(r-1), then order ${\mathrm{\&mu;}}_{\mathrm{ij}}^{\&prime;}\left(r\right)\stackrel{\mathrm{\&Delta;}}{=}{\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right),$ $r+1\&DoubleRightArrow;r,$ And return 4) to μ _Ij(r) carry out iterative computation; Otherwise export the image that strengthens for the r-1 time.

The computation complexity of above algorithm can be analyzed as follows: for the image with m * n pixel, the computation complexity of generalized fuzzy enhancement algorithms is O (m * n).

The method of extracting target cell in the step 2 adopts active contour model method (Snake method), and specific implementation may further comprise the steps:

1) structure energy model

Setting curve v (s)=[x (s) y (s)], s ∈ [0,1], the total energy meter that defines on it is shown:

E _total(v(s))＝∫ _s(E _int(v(s))+E _image(v(s))+E _con(v(s)))ds (1)

Wherein: E _Int(v (s))=(α (s) | v _s(s) | ²+ β (s) | v _Ss(s) | ²) (2)

E _image(v(s))＝w _lineE _line(v(s))+w _edgeE _edge(v(s))+w _termE _term(v(s)) (3)

E _con(v(s))＝k(x ₁-x ₂) ² (4)

E _Int(v (s)) is internal energy, expressed the more level and smooth power of curve of ordering about, and wherein the single order item has been expressed and made the littler pulling force of consecutive point distance, and second order term has been expressed the crooked rigidity power of resisting; α (s) and β (s) expression weight separately; E _Image(v (s)) is image energy, the heat input E that to be the guiding snake that obtains from image advance towards low gray scale or high grayscale position _Line=I (x, y), edge energy uses $E_{\mathrm{edge}}=-|\&dtri;I\left(v\left(s\right)\right)|$ With the terminating point of image center line and turning ENERGY E to the influence of outline line trend _TermThe weighted sum of three energy terms; w _Line, w _EdgeAnd w _TermThe weight of each component of representative image energy; E _Con(v (s)) expression attracts the elastic force of outline line to position of image, x ₁And x ₂The specified point of representing outline line and picture position respectively.If external energy is defined as

E _ext(v(s))＝E _image(v(s))+E _con(v(s)) (5)

Then total energy is

E _total(v(s))＝∫ _s(E _int(v(s))+E _ext(v(s)))ds?(6)

2) utilize the variational method that gross energy is carried out minimization, outline line is satisfied

$F_v-\frac{\&PartialD;}{\&PartialD;s}{F}_{v_s}+\frac{{\&PartialD;}^2}{{\&PartialD;s}^2}{F}_{v_{\mathrm{ss}}}=0---\left(7\right)$

3) determine the position of cell target on image by the center of curve enclosing region, this position is as the position measurement of this cell image of present frame.

Set up the cell movement kinetic model in the step 3 and comprise and set up target movement model and measurement model that wherein target movement model is

x(k+1)＝F(k)x(k)+G(k)u(k)+v(k) (8)

Measurement model is

z(k)＝H(k)x(k)+w(k) (9)

X (k) expression target cell in k motion state (position, speed) constantly, z (k) expression k image measurement constantly, F (k), G (k) and H (k) represent k state-transition matrix, control input matrix and measurement matrix constantly respectively, and v (k) and w (k) have described the stochastic system noise respectively and measured noise.

The tracking of moving target cell adopts recursion Bayesian filtering method to upgrade each target in the step 4, obtains the current state and the estimated accuracy of each target, handles by data association for many cells tracking and cell division, cell aggregation.

Recursion Bayesian filtering method and JPDA method all adopt ripe prior art, can adopt methods such as alpha-beta filtering, Kalman filtering, PF filtering as recursion Bayesian filtering method, data association adopts JPDA method (being proposed by Bar Shalom Y).

The present invention has characteristics than very noisy and disturbance, picture contrast difference according to video frequency microscopic image, the generalized fuzzy Enhancement Method that adopts medium filtering, assessment links such as fuzzy enhancing, greyscale transformation and picture quality to constitute is handled frame by frame to original view data, can improve image processing effect largely.The more common image enchancing method of the generalized fuzzy enhancement algorithms that proposes has more performance handling aspect weak contrast, the very noisy image.

The present invention starts with from the original video micro-image that obtains, and utilization Flame Image Process and analytical approach are improved picture quality, extracted cell outline and definite cell position, thereby obtain the sequential observation data.In analysis of cells kinetic characteristic and image on the basis of noise and interference characteristic, adopt dynamic system modeling and stochastic modeling method, set up the cell movement system model, by recursion Bayesian filtering and data association technology, the movement locus of pair cell carries out from motion tracking, has provided the tracking results of cell compound movement simultaneously.The present invention has complete systematicness and very strong practicality.

Embodiment

The method of video microscopic image cell automatic tracking may further comprise the steps: 1. obtain real-time cell movement image by phase microscope, the cell movement video frequency microscopic image that obtains is carried out enhancement process frame by frame; 2. the cell movement extracting target from images cell after the enhancement process; 3. set up the cell movement kinetic model; 4, the target cell of pursuit movement.

The enhancement process of video frequency microscopic image adopts generalized fuzzy enhancement process method in the step 1, and concrete steps are as follows:

(1) in order to weaken the grain noise in the image, generally speaking, in spatial domain, can reduce noise,,, therefore can adopt various forms of low-pass filtering methods to reduce noise because noise spectrum is many at high band in frequency field with neighborhood averaging.In order to avoid soft edge can remove impulsive noise and what is called " spiced salt " noise (Salt-and-pepper noise) again as far as possible, adopt median filter to carry out smothing filtering;

(2) adopt following transform method to carry out the figure image intensifying

${\mathrm{\&mu;}}_{\mathrm{ij}}=H\left(f_{\mathrm{ij}}\right)={(1+\frac{f_{\mathrm{ij}}-f_{\max }}{f_{\max }-f_{\min }})}^{F_c}$

F wherein _cBe the fuzzy behaviour parameter, f _Ij, f _Max, f _MinPixel (i, gray-scale value j), visual maximum gray scale and minimum gradation value, μ in the difference presentation image _IjExpression pixel (i, j) fuzzy membership.The utilization interative computation is to image blurring characteristic plane { μ _IjCarry out enhancement process, obtain new fuzzy characteristics plane μ ' _Ij, and carry out following inverse transformation on this basis, obtain the fuzzy gray-scale value that strengthens the back image:

$f_{\mathrm{ij}}^{\&prime;}=H^{-1}\left({\mathrm{\&mu;}}_{\mathrm{ij}}^{\&prime;}\right)=f_{\max }-(1-{{\mathrm{\&mu;}}_{\mathrm{ij}}^{\&prime;}}^{\frac{1}{F_c}})(f_{\max }-f_{\min })$

(3) image behind fuzzy the enhancing is carried out following greyscale transformation

$f_{\mathrm{ij}}^e=t\left(f_{\mathrm{ij}}^{\&prime;}\right)=\frac{f_{\max }^e-f_{\min }^e}{f_{\max }^{\&prime;}-f_{\min }^{\&prime;}}(f_{\mathrm{ij}}^{\&prime;}-f_{\min }^{\&prime;})+f_{\min }^e$

F in the formula _Ij ^eBe the gradation of image value behind the process greyscale transformation t (), f _Min ^e, f _Max ^eBe respectively maximum, the minimum value of the gradation of image after the greyscale transformation of setting, f ' _Min, f ' _MaxBe respectively fuzzy maximum, the minimum value that strengthens gradation of image, and $f_{\min }^{\&prime;}\&GreaterEqual;f_{\min }^e,$ $f_{\max }^{\&prime;}\&le;f_{\max }^e.$

(4) utilize the standard deviation weighting of gradation of image, obtain following image quality evaluation index its grey level histogram

${\mathrm{\&sigma;}}_w={\mathrm{\&Delta;f}}^{-1}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{(p_j-\stackrel{\&OverBar;}{p})}^2/N}$

σ wherein _wThe weighting standard that is image grey level histogram is poor, p _jBe pixel number shared number percent in the total pixel N of image of j gray shade scale, p is p _jMean value, Δ f is the gradation of image scope.

(5) repeat (2)-(4) step, up to picture quality evaluation index σ _wTill no longer reducing.

The method of extracting target cell in the step 2 adopts active contour model method (Snake method), this method is a kind of mechanism of the feature of positioning image from top to bottom, at first set an initial outline line (" snake "), advance to the characteristics of image direction by the constraining force promotion outline line that acts on " snake point " then, final lock onto target structure be by minimization dynamic outline line total energy integrated metric at first, specific implementation may further comprise the steps:

(1) structure of energy model

Make curve v (s)=[x (s) y (s)], s ∈ [0,1], then definition total energy thereon can be expressed as:

E _total(v(s))＝∫ _s(E _int(v(s))+E _image(v(s))+E _con(v(s)))ds (1)

Wherein: E _Int(v (s))=(α (s) | v (s) | ²+ β (s) | v _Ss(s) | ²) (2)

E _image(v(s))＝w _lineE _line(v(s))+w _edgeE _edge(v(s))+w _termE _term(v(s)) (3)

E _con(v(s))＝k(x ₁-x ₂) ² (4)

Internal energy E _Int(v (s)) expressed the more level and smooth power of curve of ordering about, and wherein the single order item has been expressed and made the littler pulling force of consecutive point distance, and second order term has been expressed the crooked rigidity power of resisting, α (s) and β (s) representative weight separately.Image energy E _Image(v (s)) is the weighted sum of three energy terms obtaining from image: the heat input that the guiding snake advances towards low gray scale or high grayscale position, E _Line=I (x, y); Edge energy is used $E_{\mathrm{edge}}=-|\&dtri;I\left(v\left(s\right)\right)|$ Expression, thus attract outline line to the image border point that high gradient mould value is arranged; E _TermThe terminating point of presentation video center line and turning are to the influence of outline line trend.w _Line, _WedgeAnd w _TermThe weight of each component of representative image energy.E _Con(v (s)) expressed and attracted outline line to the elastic force of specifying certain position, x ₁And x ₂The specified point of representing outline line and picture position respectively.If note

E _ext(v(s))＝E _image(v(s))+E _con(v(s)) (5)

Then have

E _total(v(s))＝∫ _s(E _int(v(s))+E _ext(v(s)))ds (6)

(2) calculate based on the energy minimization of the variational method

The final position of outline line can obtain by variational method.With the integral in (6) formula F (s, vs, v _Ss) replace, then the curvilinear equation of Tui Daoing should satisfy following Euler-La bright day equation:

$F_v-\frac{\&PartialD;}{\&PartialD;s}{F}_{v_s}+\frac{{\&PartialD;}^2}{{\&PartialD;s}^2}{F}_{v_{\mathrm{ss}}}=0---\left(7\right)$

(3) on the result that the active contour model method is handled, can determine the position of cell target on image by the center of calculated curve enclosing region.This position can be used as the position measurement of current frame image cell.

Because cell movement has certain randomness, therefore its motion is described and set up the method that the cell movement kinetic model adopts stochastic modeling in the step 3, comprise and set up target movement model and measurement model that wherein target movement model is

x(k+1)＝F(k)x(k)+G(k)u(k)+v(k) (8)

Measurement model is

z(k)＝H(k)x(k)+w(k) (9)

X (k) expression target cell in k motion state (position, speed) constantly, z (k) expression k image measurement constantly, F (k), G (k) and H (k) represent k state-transition matrix, control input matrix and measurement matrix constantly respectively, and v (k) and w (k) have described the stochastic system noise respectively and measured noise.

The tracking of moving target cell adopts recursion Bayesian filtering method to upgrade each target in the step 4, obtains the current state and the estimated accuracy of each target, handles by data association for many cells tracking and cell division, cell aggregation.

A plurality of cells are often arranged on the same two field picture, and the motion of these cells is irregular, and may have biological phenomenons such as the division of cell and merging.Because the interference of image capture device performance and actual environment, noise in the image and clutter grade are higher.The feasible tracking cell based on image of these factors must solve a key technical problem: data association, the promptly correct relevance of judging a plurality of metrical informations and many tracked target flight paths.In case eliminated the uncertainty in the source of measuring, just the multiple target tracking problem can be converted into a plurality of monotrack problem solvings.Can handle by tracking gate technology and stochastic modeling method for clutter, the division of cell and merging phenomenon can be equivalent to the generation and the consolidation problem of targetpath.

JPDA (JPDA) is a kind of sub-optimal algorithm that solves the data association problem.When a plurality of target approaching in the tracking measurement space, measurement may drop in the tracking gate of several objects simultaneously, has the target of common factor to form " gather " (cluster) these tracking gates, establishes that to gather the internal object number be n _t, the measurement number that falls into these target following doors is n _m, will gather with a binary correlation logic matrix $\mathrm{\&Omega;}={\left[{\mathrm{\&Omega;}}_{\mathrm{ij}}\right]}_{n_m\&times;(n_i+1)}$ Expression.Ω _Ij=1, represent that i measurement may come from j target (j=0 represents that this is measured as clutter); Otherwise Ω _IjI measurement of=0 expression can not come from j target.Satisfy following three constraint conditions a kind of measure and target between may the pairing incident be referred to as a feasible incident χ (feasible event):

Each target produces a measurement at most;

Each is measured and derives from a target at most;

Fall into candidate within the tracking gate of certain target and measure or come from this target, perhaps come from clutter, perhaps come from other targets.

Feasible incident is represented with binary correlation logic matrix Φ by the constraint condition of feasible incident as can be known, every row element sum of Φ equals 1, every column element sum equals 1 or 0 (except the 0th row).Feasible incident can be regarded as the part combination of selecting according to three constraint conditions in all mathematical combination of measuring the formation of collection and object set undetermined.Obtain the posterior probability of each feasible incident, and with all Φ _IjThe posterior probability addition of=1 feasible incident is promptly measured the posterior probability that belongs to j target for i, the weight when obtaining this this target of measurement renewal with it.Note ${\mathrm{\&tau;}}_i\left(\mathrm{\&chi;}\right)=\underset{j=1}{\overset{n_t}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_{\mathrm{ij}}$ Represent i target function of measuring among the feasible incident χ, ${\mathrm{\&delta;}}_i\left(\mathrm{\&chi;}\right)=\underset{i=1}{\overset{n_t}{\mathrm{\&Sigma;}}}{\mathrm{\&Phi;}}_{\mathrm{ij}}$ The target function of representing j target among the feasible incident χ,

Represent clutter number among the feasible incident χ.Satisfy normal distribution in the measurement that is derived from target, clutter satisfies evenly distribution, and the clutter number satisfies under the prerequisite of Poisson distribution, and the posterior probability that gets feasible incident χ generation is

Wherein first connects the normal distribution probability of taking advantage of the expression measurement to belong to realistic objective, and second company takes advantage of the detected probability of all targets of expression, the 3rd probability of taking advantage of the expression target all not to be detected, and c is a normalized factor.Measuring the posterior probability that is derived from j target for i is

${\mathrm{\&beta;}}_{k+1}^{\left(i\right)\left(j\right)}=\underset{\mathrm{\&chi;}:{\mathrm{\&Phi;}}_{\mathrm{ij}}=1}{\mathrm{\&Sigma;}}P\left\{\mathrm{\&chi;}\right|Z^{k+1}\}---\left(11\right)$

The posterior probability that j target do not produce any measurement is

${\mathrm{\&beta;}}_{k+1}^{\left(0\right)\left(j\right)}=1-\underset{i=1}{\overset{n_m}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{k+1}^{\left(i\right)\left(j\right)}---\left(12\right)$

All are effectively measured weighting must merge measurement:

$Z_{k+1}=\underset{j=1}{\overset{N_t}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{k+1}^{\left(j\right)}{z}_{k+1}^{\left(j\right)}---\left(13\right)$

Upgrade each target according to recursion Bayesian filtering method (as methods such as alpha-beta filtering, Kalman filtering, PF filtering), obtain the current state (estimator) and the estimated accuracy (dbjective state covariance matrix) of each target.

Claims (1)

1, a kind of automatic tracking method for video frequency microscopic image cell, it is characterized in that this method may further comprise the steps: (1) obtains real-time cell movement image by phase microscope, and the cell movement video frequency microscopic image that obtains is carried out enhancement process frame by frame; (2) the cell movement extracting target from images cell after the enhancement process; (3) set up the cell movement kinetic model; (4) target cell of pursuit movement;

The enhancement process of video frequency microscopic image adopts generalized fuzzy enhancement process method in the step (1), specifically may further comprise the steps:

1. pending image f is imported in initialization, and the initial value of setting iterations is r=1;

2. adopt median filter to carry out smothing filtering, with each pixel f of image f (i, j) as the center of window, with the pixel grey scale mean value of window institute overlay image as f (i, new gray-scale value f j) _Ij

3. with the new gray-scale value f of each pixel _IjCarry out the figure image intensifying according to following transform method

${\mathrm{\&mu;}}_{\mathrm{ij}}=H\left(f_{\mathrm{ij}}\right)={(1+\frac{f_{\mathrm{ij}}-f_{\max }}{f_{\max }-f_{\min }})}^{F_c}$

F wherein _cBe the fuzzy behaviour parameter, f _Ij, f _Max, f _MinPixel f in the difference presentation video (i, gray-scale value j), the maximum gray scale of image and minimum gradation value, μ _IjRemarked pixel f (i, j) fuzzy membership;

The gray-scale value f that all are new _IjCarry out iteration the r time, determine the fuzzy characteristics plane { μ of image behind the medium filtering _Ij(r) };

4. to μ _Ij(r) make following nonlinear transformation, transformation results is designated as μ ' _Ij(r);

${\mathrm{\&mu;}}_{\mathrm{ij}}^{\&prime;}\left(r\right)=T\left({\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right)\right)=\left\{\begin{array}{cc}2{\left({\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right)\right)}^2& 0\&le;{\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right)\&le;0.5\\ 1-2{(1-{\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right))}^2& 0.5<{\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right)\&le;1\end{array}\right.---\left(1\right)$

5. to μ ' _Ij(r) do inverse transformation, obtain new gray level image f ' _Ij;

6. the image behind fuzzy the enhancing is carried out following greyscale transformation

$f_{\mathrm{ij}}^e=t\left(f_{\mathrm{ij}}^{\&prime;}\right)=\frac{f_{\max }^e-f_{\min }^e}{f_{\max }^{\&prime;}-f_{\min }^{\&prime;}}(f_{\mathrm{ij}}^{\&prime;}-f_{\min }^{\&prime;})+f_{\min }^e---\left(2\right)$

F in the formula _Ij ^eBe the gradation of image value behind the process greyscale transformation t (), f _Min ^e, f _Max ^eBe respectively maximum, the minimum value of the gradation of image after the greyscale transformation of setting, f ' _Min, f ' _MaxBe respectively fuzzy maximum, the minimum value that strengthens gradation of image, and $f_{\min }^{\&prime;}\&GreaterEqual;f_{\min }^e,$ $f_{\max }^{\&prime;}\&le;f_{\max }^e;$

7. relatively strengthen the image quality in images evaluation index the r time and the r-1 time, if σ _w(r) less than σ _w(r-1), then order ${\mathrm{\&mu;}}_{\mathrm{ij}}^{\&prime;}\left(r\right)\stackrel{\mathrm{\&Delta;}}{=}{\mathrm{\&mu;}}_{\mathrm{ij}}\left(r\right),$ $r+1\&DoubleRightArrow;r,$ And return 4. to μ _Ij(r) carry out iterative computation; Otherwise export the image that strengthens for the r-1 time;

Described image quality evaluation index is to utilize the standard deviation weighting of gradation of image to its grey level histogram, the following image quality evaluation index that obtains

${\mathrm{\&sigma;}}_w=\mathrm{\&Delta;}{f}^{-1}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{(p_j-\stackrel{\&OverBar;}{p})}^2/N}$

σ wherein _wThe weighting standard that is image grey level histogram is poor, p _jBe pixel quantity shared number percent in the total pixel N of image of j gray shade scale, p is p _jMean value, Δ f is the gradation of image scope;

The method of extracting target cell in the step (2) adopts the active contour model method, and specific implementation may further comprise the steps:

1. construct energy model

Setting curve v (s)=[x (s) y (s)], s ∈ [0,1], the total energy meter that defines on it is shown:

E _total(v(s))＝∫ _s(E _int(v(s))+E _image(v(s))+E _con(v(s)))ds (3)

Wherein:

E _int(v(s))＝(α(s)|v _s(s)| ²+β(s)|v _ss(s)| ²) (4)

E _image(v(s))＝w _lineE _line(v(s))+w _edgeE _edge(v(s))+w _termE _term(v(s))?(5)

E _con(v(s))＝k(x ₁-x ₂) ² (6)

E _Int(v (s)) is internal energy, expressed the more level and smooth power of curve of ordering about, and wherein the single order item has been expressed and made the littler pulling force of consecutive point distance, and second order term has been expressed the crooked rigidity power of resisting; α (s) and β (s) expression weight separately; E _Image(v (s)) is image energy, the heat input E that to be the guiding snake that obtains from image advance towards low gray scale or high grayscale position _Line=I (x, y), edge energy $E_{\mathrm{edge}}=-|\&dtri;I\left(v\left(s\right)\right)|$ With the terminating point of image center line and turning ENERGY E to the influence of outline line trend _TermThe weighted sum of three energy terms; w _Line, w _EdgeAnd w _TermThe weight of each component of representative image energy; E _Con(v (s)) expression attracts the elastic force of outline line to the picture position, x ₁And x ₂The specified point of representing outline line and picture position respectively;

2. utilize the variational method that gross energy is carried out minimization, outline line is satisfied

$F_v-\frac{\&PartialD;}{\&PartialD;s}{F}_{v_s}+\frac{{\&PartialD;}^2}{\&PartialD;s^2}{F}_{v_{\mathrm{ss}}}=0---\left(7\right)$

3. determine the position of cell target on image by the center of curve enclosing region, this position is as the position measurement of this cell image of present frame;

Set up the cell movement kinetic model in the step (3) and comprise and set up target movement model and measurement model that wherein target movement model is

x(k+1)＝F(k)x(k)+G(k)u(k)+v(k) (8)

Measurement model is

z(k)＝H(k)x(k)+w(k) (9)

X (k) expression target cell in k motion state constantly, z (k) expression k image measurement constantly, F (k), G (k) and H (k) represent k state-transition matrix, control input matrix and measurement matrix constantly respectively, and v (k) and w (k) have described the stochastic system noise respectively and measured noise;

The tracking of moving target cell adopts recursion Bayesian filtering method to upgrade each target in the step (4), obtains the current state and the estimated accuracy of each target, handles by data association for many cells tracking and cell division, cell aggregation.