CN100535943C - High light hot spot eliminating method using for visual convex shell drawing and device thereof - Google Patents

Abstract

This invention relates to an elimination method for high brightness spots and its device used in drawing visual convex hulls, which eliminates high brightness spots in an image to get non-high brightness image sequence finally with the help of calibration information of an image collection device and redundant information of image sequences in drawing visual convex hulls including the following steps: picking up images of the high brightness spot, searching for corresponding sub-image in other reference image and re-sampling the image to further process the re-sampled sub-image seamlessly. This invention also provides a related hardware device.

Description

A kind of high light hot spot eliminating method using and device thereof that is used for visual convex shell drawing

Technical field

The invention belongs to computer graphics and virtual reality technology field, be specifically related to a kind of high light hot spot eliminating method using and device thereof that vision convex hull (Visual Hull) is drawn that be used for.

Background technology

An important goal of 3D computer graphics is exactly in computing machine three-dimensional scenic to be drawn out realistic image.Traditional method for drafting based on geometric model becomes basic geometric primitive with scene description, defines distribution and character, body surface material and the reflection characteristic of light source then, produces virtual composograph by the calculating of drawing formula at last.This method for drafting exists inevitable problem: the object in the reality and the complexity of scene are difficult to describe accurately with the simple geometric element, and the increase of scene complexity can make the calculated amount of drafting obviously increase.

The rendering technique based on image of Chu Xianing (Image-based Rendering in recent years, IBR) scene of directly utilize gathering or subject image are as input, be not subjected to the influence of its complexity, therefore the incomparable advantage of traditional rendering technique is arranged when drawing actual scene.As, document " Image-Based Visual Hull " (Matusik, W., Buehler, C., et al.Proceedings of ACM SIGGRAPH 2000,369-374,2000.) has proposed a kind of vision convex hull algorithm based on image.

In vision convex hull algorithm, at first object is gathered some reference pictures as shown in Figure 1, and foreground object and background are cut apart based on image.The bianry image that obtains after cutting apart is called outline profile (silhouetteimage), and the ray that is cast out by the camera center and the common factor of silhouette are called the projection awl.As shown in Figure 2, the common factor of the projection of all reference pictures awl promptly is a kind of approximate expression to the true three-dimension solid object surface, is called vision convex hull.Theoretically, reference picture is many more, and vision convex hull is approaching more real body surface just.This method only need be known camera calibration information and the objective contour when taking, and this is easier to try to achieve in practice.

But,, thereby bring harmful effect for the final drafting effect of object because can there be high bright spot in the condition restriction of images acquired inevitably in the image that collects.As: the high bright spot on the reference picture too much can be covered the texture of object itself, and high bright spot is obvious more, and the texture of new view is also just untrue more.

Eliminate this harmful effect, just need the high bright spot in the reference picture be removed by means such as image repair.Existing high light elimination algorithm generally has specific (special) requirements to input picture, as: must synthesize the image of no Gao Guang by analyzing the difference that compares two width of cloth images then at each width of cloth of image of identical station acquisition flash illumination and ambient light illumination.Another kind of method can be eliminated Gao Guang on the single image by revising the image gradient territory, but generally can only obtain effect preferably to the more single zone of background.In the drawing process of vision convex hull, the reference image sequence that is used to generate convex hull often reaches tens of width of cloth images, and the color of object, texture be also more complicated usually, so said method all is difficult to be suitable for.

Summary of the invention

Purpose of the present invention promptly is the deficiency at existing method, proposes a kind of high light hot spot eliminating method using and device thereof that is applicable to vision convex hull calculating, drawing process.This method is utilized camera calibration information in the vision convex hull generative process and the redundant information between the reference picture, and the highlight area of every width of cloth image is revised respectively, finally obtains not having the image sequence of high bright spot.Use revised image sequence to carry out visual convex shell drawing, the harmful effect that can effectively avoid high light that drawing result is brought.

For achieving the above object, high light hot spot eliminating method using of the present invention mainly is divided into following step (as shown in Figure 3):

1) extracts high bright spot subimage

At each width of cloth reference picture that image collecting device obtains, choose the subimage that wherein comprises high bright spot.

2) seek corresponding subimage

To choosing the subimage that comprises high bright spot in the 1st step, utilize the image collecting device calibration information to shine upon, in other reference pictures, find its corresponding subimage respectively.

3) the bright spot subimage of height is resampled

Utilize the colouring information of each corresponding subimage that selected high bright spot subimage is resampled, revise the color of high light pixel.

Above-mentioned method further comprises the steps:

4) the subimage edge after resampling is carried out seamless process and handle, eliminate the non-continuous event at subimage edge.

The present invention also provides a kind of figure image height light spot cancellation element, comprising:

The sequence image harvester is used to gather reference image sequence, can be that digital camera or CCD digital camera are first-class;

High bright spot subimage extraction element: be used for extracting the high bright spot subimage that comprises high bright spot from each width of cloth reference picture;

Corresponding subimage mapping device: be used for inside and outside parameter, in other reference pictures, find the corresponding subimage of the bright spot subimage of current height according to the sequence image harvester;

High bright spot subimage resampling device is used to utilize the colouring information of the corresponding subimage on other reference pictures that the bright spot subimage of current height is resampled, and eliminates high bright spot.

Above-mentioned device further comprises subimage edge seamless process treating apparatus, is used to eliminate the color non-continuous event on the sub-image boundary after resampling.Can also comprise the sequence image output unit, be used to export the sequence image of having eliminated high bright spot.

The invention has the advantages that and utilized the image collecting device calibration information and the contour of object information that provide in the visual convex shell drawing process to eliminate high bright spot, calculate simple and effective.Input image sequence is not had other specific (special) requirements, and the image that uses ordinary digital camera or CCD digital camera to obtain under any light source gets final product, and it is right to need not at each station acquisition image, to texture comparatively complex image also can be suitable for.Handle by further seamless process, can make the subimage and the seamless fusion of original image of resampling, obtain the no high light compositing image of high realism.

Description of drawings

Fig. 1 is the synoptic diagram that sequence image is gathered in the vision convex hull generative process;

Fig. 2 is a vision convex hull generating principle synoptic diagram;

Fig. 3 is the schematic flow sheet of high light hot spot eliminating method using;

Fig. 4 is a synoptic diagram of choosing high bright spot subimage on piece image;

Fig. 5 is a principle schematic of seeking respective pixel on adjacent image;

Fig. 6 is the example that finds high bright spot subimage corresponding region on adjacent image;

Fig. 7 resamples by subimage to eliminate the synoptic diagram of high bright spot;

Fig. 8 is the example of color non-continuous event on the resampling subimage edge;

Fig. 9 is the guide field principle schematic in the image;

Figure 10 carries out the subimage example that seamless process is handled by finding the solution Poisson equation;

Figure 11 (a) and (b) are examples as a result that high bright spot is eliminated, and the left side is an original image among the figure, and the right side is the image after too high light is eliminated;

Figure 11 (c) is to use revised sequence image to carry out the example as a result of visual convex shell drawing, and the left side is the result who directly draws with original image, and the right side is to use the image sequence of eliminating Gao Guang to carry out the new view of visual convex shell drawing;

Figure 12 is a figure image height light spot cancellation element structural representation.

Embodiment

Following with reference to accompanying drawing, describe specific embodiments example of the present invention in detail.

As shown in Figure 1, based on the vision convex hull method of image, being input as of its drawing process centers on one group of reference picture that object is taken.Input image sequence should have following characteristics:

1. the locus ordering of camera when all images is by shooting;

2. per two width of cloth adjacent images should have overlapped part;

3. the entire image sequence can comprise each orientation of object.

By the method for camera calibration, can obtain the inside and outside parameter of each reference picture camera.With methods such as image segmentation, foreground object and background can be separated again, obtain the two-dimensional silhouette of object on each reference picture.Utilize above these information, can generate the vision convex hull model of object, and draw out synthetic image from arbitrarily new visual angle.

Eliminate the high bright spot in the reference picture, at first need from every width of cloth image, to extract the subimage that comprises highlight area.This can realize by manual interaction or automatic Calculation Method.Fig. 4 is exactly an example of having chosen high bright spot subimage.

After having extracted high bright spot subimage, respectively each subimage is calculated, utilize the colouring information of correspondence on the contiguous reference picture that the bright spot subimage of current height is resampled, thereby the bright spot of height is eliminated.Therefore at first need to find the corresponding region of current subimage on contiguous reference picture.Need to know the inside and outside parameter of camera in this step, this is provided by the camera calibration algorithm in the generative process of vision convex hull.

As shown in Figure 5, known image I ₀On 1 P ₀, reference picture I _k, camera parameter is known, calculate P ₀In image I _kOn corresponding point P _kThe concrete steps of calculating are as follows:

1. from current camera C ₀Set out to construct and pass P in the center ₀Ray r;

2. because camera is demarcated, we at first by the calibrating parameters of camera, obtain I ₀With I _kFundamental matrix (fundamental matrix), calculate P ₀Point is in image I _kOn polar curve (epipolarline) l _e, the intersection point of it and contour of object is respectively P _a ^kAnd P _b ^k, from the camera center C ₀The ray C that sends ₀P ₀Respectively with C _kP _a ^k, C _kP _b ^kIntersect (according to polar curve how much, these three rays are positioned at same plane), the line segment note that intersection point constitutes is made V _a ^kV _b ^k

3. repeating step 2, until handling all effective reference pictures, obtain one group of line segment { V _a ^kV _b ^k| k=1 ..., n};

4. this group line segment is found the solution its maximum public sub-line segment, note is made V _aV _b

5. with line segment V _aV _bGo up the V nearer apart from viewpoint _aSpot projection is to plane of delineation I _k, obtain corresponding point P _k

Like this, just obtained the respective pixel of arbitrary pixel on adjacent image in the high bright spot subimage.Each pixel in the subimage is repeated above-mentioned steps, can obtain the corresponding region of high bright spot subimage on its contiguous reference picture, the result as shown in Figure 6.

Because photoenvironment is relatively stable, the angle difference of shot object, certain displacement can appear in the highlight area in the image sequence.For example, only some is a highlight area to the high bright spot of the left figure of Fig. 6 in middle figure, and therefore complete obiteration on the right figure can suppose that most of pixel of corresponding region is non-highlight.On this basis, for the arbitrary pixel P in the highlight area ₀, can suppose its respective pixel set { p _k| k=1 ... most of color of pixel is near true colors among the n}.So, give certain weight w for each respective pixel _kAfter mix, just can synthesize new pixel near realistic colour.

The color that target object reflects under the illumination is made up of diffuse reflection (diffuse) component and direct reflection (specular) component.For non-highlight area, the pixel of each reference picture is mainly by the decision of diffusion component, thereby color distribution is concentrated; And in the highlight area, color distribution relative departure.Therefore, in pixel mixed process, a reasonable weight value distribution principle is: depart from the pixel of big more (variance is big more) for color distribution, the weight that should give is low more.

In actual applications, can determine the weight of each pixel briefly with gray-value image.Suppose g _kBe pixel p _kCorresponding gray scale value, then { p _k| k=0 ... the average gray mean value of n} is

$g_{\mathrm{avg}}=\frac{1}{n}\underset{k=0}{\overset{n-1}{\mathrm{\Σ}}}{g}_k.$

According to the gray-scale value variance, it is as follows to define pixel weight:

$w_k=c\frac{1}{{|g_{\mathrm{avg}}-g_k|}^2},$

Wherein c is a constant, to guarantee w _kSum equals 1.So the new pixel rgb value that mixing obtains through pixel is:

$\left(\begin{array}{c}R\\ G\\ B\end{array}\right)=\left(\begin{array}{c}\underset{k=0}{\overset{n}{\mathrm{\Σ}}}{w}_k{R}_k\\ \underset{k=0}{\overset{n}{\mathrm{\Σ}}}{w}_k{G}_k\\ \underset{k=0}{\overset{n}{\mathrm{\Σ}}}{w}_k{B}_k\end{array}\right),$

(P wherein _k, G _k, B _k) be p _kThe RGB component.

In general, the image of employing is many more, the respective pixel that drops on highlight area so comparatively speaking still less, therefore the color that obtains is also more approaching with reality.But, ask for corresponding point and might produce site error, so partly be easy to generate blooming in complex texture owing to the numerical error that camera calibration, three-dimensional computations etc. are brought.Therefore, in statistical pixel, also be provided with a threshold tau, control the quantity of choosing picture with it.Suppose that the subimage of choosing comprises m pixel, the gray-scale value of each pixel is b _i, the average gray of this area pixel is so:

$b_{\mathrm{avg}}=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{b}_i,$

Gray variance is

$b_{\mathrm{dev}}=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|b_i-b_{\mathrm{avg}}|}^2.$

So number of reference pictures (also promptly selecting the quantity of picture in the pixel mixed process for use) n that determines is

$n=\left\{\begin{array}{cc}{N}_{\max },& b_{\mathrm{dev}}<\mathrm{\&tau;}\\ N_{\min },& b_{\mathrm{dev}}\&GreaterEqual;\mathrm{\&tau;}\end{array}\right..$

V wherein _MaxAnd N _MinBe empirical value, the meaning of choosing picture number like this is: for the big more highlight area of variance, texture is complicated more, therefore should select for use less number of reference pictures to prevent to blur; For the comparatively simple zone of texture, can select more reference picture for use, thereby obtain more level and smooth effect.During concrete enforcement, will from all reference pictures, choose shooting angle and the immediate n width of cloth of target image participates in computing.Be exactly to adopt said method to remove the result of Gao Guang among Fig. 7, the left side is former figure, and wherein rectangle is high light constituency; The right side is revised image.

The method that adopts pixel to mix, replace can be removed the Gao Guang of subimage inside preferably, but in some cases, certain non-continuous event appears in subimage edge part branch.Background color as white among Fig. 8 is comparatively responsive, therefore has the jump of color on the edge.For addressing this problem, the present invention is by finding the solution the gradient field that Poisson equation comes correction image, Laplace operator on the subimage field of definition is consistent with the subimage of eliminating Gao Guang, and on the border, is consistent, obtain level and smooth boundary effect at last with original image.

As shown in Figure 9, suppose that entire image field of definition I is R ²On a closed subset, Ω is a subclass on the I,

Be this regional border, suppose the scalar function f on the I-Ω of known definition territory ^*, other has one to be defined in vector field v on the Ω (x y), is referred to as guide field (guidance field).The target of finding the solution is the scalar function f that obtains on the Ω, thereby makes this regional gradient field

With guide field v (x, y) unanimity.

Because this vector field non-conservation can not directly be found the solution by the method for integration.So from all two-dimensional functions, its gradient field of searching and guide field v under the meaning of least square method (this problem is equivalent to find the solution the minimum value of following integration for x, y) immediate separating:

$\min \&Integral;\&Integral;F(\&dtri;f,v)\mathrm{dxdy},$ Wherein $f{|}_{\&PartialD;\mathrm{\&Omega;}}=f^{*}{|}_{\&PartialD;\mathrm{\&Omega;}}.$

In the following formula,

$F(\&dtri;f,v)={\left|\right|\&dtri;f-v\left|\right|}^2={(\frac{\&PartialD;f}{\&PartialD;x}-v_x)}^2+{(\frac{\&PartialD;f}{\&PartialD;y}-v_y)}^2.$

According to variational principle, the minimum value of above-mentioned integration satisfies Euler-Lagrange's equation:

$\frac{\&PartialD;F}{\&PartialD;f}-\frac{d}{\mathrm{dx}}\frac{\&PartialD;F}{\&PartialD;f_x}-\frac{d}{\mathrm{dy}}\frac{\&PartialD;F}{{\&PartialD;f}_y}=0$

Bring F into the following formula differential equation, obtain:

$2(\frac{{\&PartialD;}^{2}f}{{\&PartialD;x}^2}-\frac{\&PartialD;v_x}{\&PartialD;x})+2(\frac{{\&PartialD;}^{2}f}{\&PartialD;y^2}-\frac{{\&PartialD;v}_y}{\&PartialD;y})=0$

The abbreviation following formula promptly gets Poisson equation:

Δf＝div?v。

Wherein Δ is a Laplace operator: $\mathrm{\&Delta;f}=\frac{{\&PartialD;}^{2}f}{\&PartialD;x^2}+\frac{{\&PartialD;}^{2}f}{\&PartialD;y^2},$

Div v be vector field v (x, divergence y): $\mathrm{div}v=\frac{\&PartialD;v_x}{\&PartialD;x}+\frac{\&PartialD;v_y}{\&PartialD;y}.$

For finding the solution above-mentioned Poisson equation, at first to determine

Border condition.In general can think Be the Neumann border.Can estimate the Laplace operator Δ in order to following equation:

Δ f (x, y) ≈ f (x+1, y)+f (x-1, y)+f (x+1, y+1)+f (x, y-1)-4f (x, y) estimate divergence operator div with the method for backward difference:

div?v≈v _x(x，y)-v _x(x-1，y)+v _y(x，y)-v _y(x-1，y)

Because Laplace operator Δ and divergence operator div are all linear operator, find the solution system of linear equations and come approximate evaluation for therefore available one.The numerical solution of Poisson equation has a variety of, in order to improve efficiency of algorithm, adopts over-relaxation iterative method to find the solution among the present invention.Over-relaxation iterative method is the method for a kind of accelerating convergence of Gauss-Seidel process of iteration, also is one of effective solution of large-scale sparse matrix system of linear equations, and it utilizes the neighbours residence and the divergence value of current pixel that current pixel is carried out iteration.Suitably control the iterations of this algorithm, just can comparatively accurately approach the ideal solution of Poisson equation.Its pseudo-code is as follows:

for?each?iterative?count

for?each?pixel(x，y)in?the?region

if?mask(x，y)is?true

then?newPixel(x，y)←[Sum?of?oldPixel(x，y)’s?4neighbors-divG(x，y)]/4*Omiga

+oldPixel(x，y)*(1-Omiga)；

For the seamless problem of duplicating among the present invention, key is will guarantee the selection border color continuous and level and smooth.Therefore, basic a selection to the guide field is exactly

$v=\&dtri;g,$

Wherein g is for mixing the subimage of having eliminated Gao Guang through pixel.From the definition of each operator, we have:

Therefore, Poisson equation Δ f=div v can abbreviation be:

Δ f=Δ g, wherein $f{|}_{\&PartialD;\mathrm{\&Omega;}}=f^{*}{|}_{\&PartialD;\mathrm{\&Omega;}}.$

Use over-relaxation iterative method to find the solution following formula, just can obtain the approximate solution of Poisson equation apace, promptly revised subimage.

To the vase among Fig. 8 adopt said method carry out seamless spliced after, the result who obtains is as shown in figure 10.Can see, after employing Poisson equation method has kept the border in constituency, the phenomenon that color is jumped does not take place at the selection border place, so the result be also more natural.

Use above-mentioned high light elimination algorithm, can repair respectively, remove high bright spot, shown in Figure 11 (a) and (b) the reference image sequence of vision convex hull.In the visual convex shell drawing process, use the reference picture after repairing to carry out the drafting of new viewpoint, just can obtain not having the new view (Figure 11 (c)) of Gao Guang.Can see that because the influence of having removed Gao Guang, new drawing result has the better sense of reality.

One skilled in the art will appreciate that image high light hot spot eliminating method using proposed by the invention not only can realize with software, can realize with hardware equally.As shown in figure 12, figure image height light elimination hardware unit can comprise:

A kind of figure image height light spot cancellation element comprises:

The sequence image harvester: being used to gather reference image sequence, can be that digital camera or CCD digital camera are first-class;

High bright spot subimage extraction element: be used for extracting the high bright spot subimage that comprises high bright spot from each width of cloth reference picture;

Corresponding subimage mapping device: be used for inside and outside parameter, in other reference pictures, find the corresponding subimage of the bright spot subimage of current height according to the sequence image harvester;

High bright spot subimage resampling device: be used to utilize the colouring information of the corresponding subimage on other reference pictures that the bright spot subimage of current height is resampled, eliminate high bright spot;

Subimage edge seamless process treating apparatus: be used to eliminate the color non-continuous event on the sub-image boundary after resampling;

Sequence image output unit: be used to export the sequence image of having eliminated high bright spot.

And then hardware unit is got in this height light spot elimination can be in conjunction with the visual convex shell drawing system of existing visual convex shell drawing hardware formation.

Although disclose specific embodiments of the invention and accompanying drawing for the purpose of illustration, its purpose is to help to understand content of the present invention and implement according to this, but it will be appreciated by those skilled in the art that: without departing from the spirit and scope of the invention and the appended claims, various replacements, variation and modification all are possible.Therefore, the present invention should not be limited to most preferred embodiment and the disclosed content of accompanying drawing.

Claims (13)

1. a high light hot spot eliminating method using that is used for the visual convex shell drawing process is characterized in that, may further comprise the steps:

(1) extracts high bright spot subimage:, choose the subimage that wherein comprises high bright spot at each width of cloth reference picture that image collecting device obtains;

(2) seek corresponding subimage: to the subimage that comprises high bright spot that step (1) is chosen, utilize the calibration information of image collecting device to shine upon, find its corresponding subimage respectively in other reference pictures, the method for specifically obtaining corresponding subimage is:

Known subimage I ₀On known point P ₀, reference picture I _k, the image collecting device parameter is known, adopts following method to draw known point P ₀At reference picture I _kOn corresponding point P _k:

A. from present image harvester center C ₀The structure that sets out passes known point P ₀Ray r;

B. by the calibrating parameters of image collecting device, obtain known subimage I ₀With reference picture I _kFundamental matrix, calculate known point P ₀At reference picture I _kOn polar curve l _e, and polar curve l _eIntersection point P with contour of object _a ^kAnd P _b ^k, from the image collecting device center C ₀The ray C that sends ₀P ₀Respectively with ray C _kP _a ^k, C _kP _b ^kIntersect, determine the line segment V that intersection point constitutes _a ^kV _b ^k

C. repeating step b until handling all reference pictures, obtains one group of line segment { V _a ^kV _b ^k| k=1 ..., n}, n are the number of reference picture;

D. this group line segment is found the solution its maximum public sub-line segment V _aV _b

E. with line segment V _aV _bGo up the end points V nearer apart from viewpoint _aBe projected to the reference picture planar I _k, obtain corresponding point P _k

To known subimage I ₀In each pixel repeat above-mentioned steps, can obtain the corresponding subimage of high bright spot subimage on its contiguous reference picture;

(3) the bright spot subimage of height is resampled: utilize the colouring information of the corresponding subimage that step (2) obtains that the selected high bright spot subimage of step (1) is resampled, revise the color of high light pixel.

2. the method for claim 1 is characterized in that, following steps are continued in described step (3) back:

(4) the subimage edge after described step (3) resampling is carried out seamless process and handle, eliminate the non-continuous event at subimage edge.

3. the method for claim 1 is characterized in that, extracts high bright spot subimage in the described step (1) and realizes by manual interaction or automatic Calculation Method.

4. the method for claim 1, it is characterized in that, seek the current corresponding subimage of subimage on other reference pictures that comprises high bright spot in the described step (2), be to utilize the image collecting device inside and outside parameter that provides by the image collecting device calibration algorithm in the vision convex hull generative process, calculate and draw through geometric transformation.

5. the method for claim 1 is characterized in that, when described step (2) is sought corresponding subimage, only finds corresponding subimage respectively in contiguous reference picture.

6. the method for claim 1 is characterized in that, revises the color of high light pixel in the described step (3) as follows: for the arbitrary pixel P in the highlight area ₀, to its respective pixel set { p _k| k=1 ... each respective pixel is given certain weight w among the n} _kAfter mix synthetic new pixel near realistic colour.

7. method as claimed in claim 6 is characterized in that the weight of described each pixel adopts the gray-value image decision as follows:

For pixel p _kCorresponding gray scale value g _k, { p _k| k=0 ... the average gray mean value of n} is:

$g_{\mathrm{avg}}=\frac{1}{n}\underset{k=0}{\overset{n-1}{\mathrm{\&Sigma;}}}{g}_k,$

According to the gray-scale value variance, it is as follows to define pixel weight:

$w_k=c\frac{1}{{|g_{\mathrm{avg}}-g_k|}^2},$

Wherein c is a constant, to guarantee w _kSum equals 1.

8. the method for claim 1 is characterized in that, controls the quantity of reference picture in the described step (2) as follows:

If the subimage of choosing comprises m pixel, the gray-scale value of each pixel is b _i, the average gray of this area pixel is so

$b_{\mathrm{avg}}=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{b}_i,$

Gray variance is

$b_{\mathrm{dev}}=\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{|b_i-b_{\mathrm{avg}}|}^2,$

So the quantity n of reference picture is

$\left\{\begin{array}{cc}{N}_{\max },& b_{\mathrm{dev}}<\mathrm{\&tau;}\\ N_{\min },& b_{\mathrm{dev}}\&GreaterEqual;\mathrm{\&tau;}\end{array}\right.,$ Wherein τ is a threshold value that is provided with, N _MaxAnd N _MinBe empirical value.

9. method as claimed in claim 2, it is characterized in that, it is by finding the solution the gradient field that Poisson equation comes correction image that seamless process in the described step (4) is handled, Laplace operator on the subimage field of definition is consistent with the subimage of eliminating Gao Guang, and on the border, be consistent with original image, obtain level and smooth boundary effect at last.

10. figure image height light spot cancellation element comprises:

Sequence image harvester: be used to gather reference image sequence;

High bright spot subimage extraction element: be used for extracting the high bright spot subimage that comprises high bright spot from each width of cloth reference picture;

Corresponding subimage mapping device: be used for inside and outside parameter, in other reference pictures, find the corresponding subimage of the bright spot subimage of current height according to the sequence image harvester

High bright spot subimage resampling device: be used to utilize the colouring information of the corresponding subimage on other reference pictures that the bright spot subimage of current height is resampled, eliminate high bright spot;

Described corresponding subimage mapping device obtains corresponding subimage as follows:

Known subimage I ₀On known point P ₀, reference picture I _k, the image collecting device parameter is known, adopts following method to draw known point P ₀At reference picture I _kOn corresponding point P _k:

A. from current sequence image collecting device center C ₀The structure that sets out passes known point P ₀Ray r;

B. by the calibrating parameters of sequence image harvester, obtain known subimage I ₀With reference picture I _kFundamental matrix, calculate known point P ₀At reference picture I _kOn polar curve l _e, and polar curve l _eIntersection point P with contour of object _a ^kAnd P _b ^k, from sequence image harvester center C ₀The ray C that sends ₀P ₀Respectively with ray C _kP _a ^k, C _kP _b ^kIntersect, determine the line segment V that intersection point constitutes _a ^kV _b ^k

C. repeating step b until handling all reference pictures, obtains one group of line segment { V _a ^kV _b ^k| k=1 ..., n}, n are the number of reference picture;

D. this group line segment is found the solution its maximum public sub-line segment V _aV _b

E. with line segment V _aV _bGo up the end points V nearer apart from viewpoint _aBe projected to the reference picture planar I _k, obtain corresponding point P _k

To known subimage I ₀In each pixel repeat above-mentioned steps, can obtain the corresponding subimage of high bright spot subimage on its contiguous reference picture.

11. device as claimed in claim 10 is characterized in that also comprising subimage edge seamless process treating apparatus, is used to eliminate the color non-continuous event on the sub-image boundary after resampling.

12. device as claimed in claim 10 is characterized in that also comprising the sequence image output unit, is used to export the sequence image of having eliminated high bright spot.

13., it is characterized in that described sequence image harvester is digital camera or CCD digital camera as claim 10 or 11 or 12 described devices.

Images