CN103093447A - Cutting and splicing method of concentration of pictures of computer - Google Patents

Abstract

The invention discloses a cutting and splicing method of concentration of pictures of a computer. The cutting and splicing method comprises the following steps: 1, estimating an importance degree; 2, dividing sub regions of a display panel to obtain a display of each picture: selecting a group of circles which are the same with the concentration pictures in the number, wherein proportion among initial radiuses of the circles and proportion among the important degrees of each picture are the same, filling the display panel through a circle filling method, after filling is finished, obtaining the sub regions of the display panel through division of circumscribed polygons of the circles, and accordingly confirming a display space Qp, on the display panel, of each picture; 3, optimizing display parameters: determining an orientation angle theta, a putting position and a scaling size of each picture; 4, carrying out seamless fusion rendering. The cutting and splicing method has the advantages of well satisfying objective needs of people on people pictures. Through arrangement of face detection, the displaying of faces is a highest priority, and accordingly the situation that the faces of the people are shielded by display regions of other pictures is avoided effectively.

Description

A kind of computing machine picture concentrates picture to shear joining method

Technical field

The present invention relates to a kind of computing machine picture and concentrate picture to shear joining method, belong to the fields such as computer picture, multimedia information technology.

Background technology

Day by day universal along with hand-held camera installation, people take pictures convenient, and the picture obtained is abundanter, then the demand of the automatic administrative skill of pictures is highlighted more.As a kind of important pictures summary and display technique, picture is pieced together and just is being subject in recent years increasing attention.The fundamental purpose that picture is pieced together be to provide one compact, abundant and attractive in appearance pictures summary figure.Such piece summary figure together need to use some professional picture editor's skills simultaneously consuming time a lot of owing to manually making, the summary figure generation technique of piecing together of robotization becomes a study hotspot in recent years.

Traditional picture is pieced method together generally based on an Optimization Framework of structure, and the quality of piecing result together is quantized with certain objective standard, and this optimization method has complicated non-linear form usually.Due to the Determines of every pictures, in some parameters, altogether may produce hundreds and thousands of parameters needs to optimize.The optimization solution that solves scale like this in so complicated nonlinear optimization equation is poor efficiency normally, also is easy to be absorbed in locally optimal solution simultaneously.Although some methods that promote solution efficiency successively are suggested, these class methods still have its intrinsic limitation.On the one hand, the optimum state parameter of every pictures often with the state tight coupling of some other pictures, cause the renewal of certain pictures state of overall importance or have influence on partly other picture, thereby cause the extensibility pieced together very poor.On the other hand, each pictures is all unified to treat usually, and the zoom factor of picture seldom is treated as important influence factor, and if the importance degree information of different pictures is added to consideration, better pictures summary can be provided, contribute to the user to control the final result of piecing together according to the hobby of oneself simultaneously.

Summary of the invention

Goal of the invention: the invention provides picture in a kind of computing machine and shear joining method, can give prominence to the personage's main body in each picture, more completely retain the core information of picture.

Technical scheme: the invention discloses in a kind of computing machine picture and shear joining method, its core be to determine the display space of picture in final synthetic picture with and displaying contents, comprise the following steps:

Step 1, importance degree assessment: weigh the color characteristic of every width picture in pictures, obtain the color complexity S of every width picture by the diversity of color in the statistics picture _c; EMD distance between employing cubic metre of earth concentrated each picture of displacement method calculating picture (EarthMover ' s Distance, be called for short the EMD distance), the similarity S using the EMD of every width picture and other pictures apart from minimum value other pictures in this picture and pictures _d, by S _cand S _dweighted sum as the importance degree S of this picture _i.

Step 2, the display panel subregion is divided the display space that obtains every width picture: choose the one group circle identical with picture number in pictures, wherein the ratio between the importance degree of ratio and the described every width picture between initial radium of each circle is identical, fill display panel by circle fill method (Circle Packing), divide and obtain the display panel subregion by the circumscribed polygon of circle after filling completes, thereby determine the display space Q of every width picture on display panel _p.

Step 3, display parameter optimization: determine every width picture towards angle θ, putting position, convergent-divergent yardstick, so that each width picture presents substantially its important area in the limited panel subregion space obtained, obtain the displaying contents of picture on panel.At first in given range, set each width picture towards angle θ; Calculate the importance degree figure (Saliency Map) of each picture, for meaning the significance level of picture pixel.To contain in the picture of facial image the importance value of each pixel in the facial image zone and be set to maximal value; Pixel higher than default importance degree threshold value in each picture is formed to connected region, after sorting from large to small according to the connected region area, the connected region of area sequence front 1/3 is divided in a polygonal region, and this polygonal region is the ROI zone Q of this picture _r(Region of Interest is called for short the ROI region-of-interest).Then the zone of the ROI towards angle θ, the picture Q set according to picture _rand definite display space Q corresponding to picture in step 2 _p, determine picture putting position and the convergent-divergent yardstick of information loss minimum.

Step 4, seamless blended is played up, and the Based on Probability mixture model carries out seamless blended to the borderline region between the display panel subregion to be played up, thereby complete the computing machine picture, concentrates picture to shear splicing.

In step 1, the importance of passing judgment on picture by complexity and the identification of picture.

Described step 1 specifically comprises the following steps:

Step 11, for weighing the complexity of picture, this method is chosen the simplest color characteristic and is weighed.This method HSV(H, Hue, tone passage; S, Saturation, saturation degree; V, Value, brightness) complexity of statistics with histogram information definition color on color space.Color complexity on the H passage is defined as follows:

$S_c^H=1-\frac{\sqrt{{\mathrm{\Σ}}_{i=1}^{m^H}{(h_i^H-\frac{1}{m^H})}^2}}{{\mathrm{\δ}}_{\max }^H}---\left(1\right);$

M wherein ^hrepresent the number of partitions of statistic histogram on the H passage, in this paper experiment, be made as 16;

be defined as and drop on pixel frequency in i subregion; for the maximum magnitude of all frequencies, as color complexity S _cnormalized factor.If the H passage of picture is constant, the histogram of H passage only has a subregion that value is arranged, at this moment

be defined as

${\mathrm{\δ}}_{\max }^H=\sqrt{{(1-\frac{1}{m^H})}^2+\frac{m^H-1}{{\left(m^H\right)}^2}}---\left(2\right);$

Complexity value on the S passage

adopt following formula to calculate:

$S_c^S=1-\frac{\sqrt{{\mathrm{\Σ}}_{i=1}^{m^S}{(h_i^S-\frac{1}{m^S})}^2}}{{\mathrm{\δ}}_{\max }^S}---\left(3\right);$

M wherein ^sfor the number of partitions of statistic histogram on the S passage, in this paper experiment, be made as 16;

be defined as and drop on pixel frequency in i subregion;

for the maximum magnitude of all frequencies, as color complexity normalized factor; If the S passage of picture is constant, be defined as and adopt following formula to calculate:

${\mathrm{\δ}}_{\max }^S=\sqrt{{(1-\frac{1}{m^S})}^2+\frac{m^S-1}{{\left(m^S\right)}^2}}---\left(4\right);$

Complexity value on the V passage

adopt following formula to calculate:

$S_c^V=1-\frac{\sqrt{{\mathrm{\Σ}}_{i=1}^{m^V}{(h_i^V-\frac{1}{m^V})}^2}}{{\mathrm{\δ}}_{\max }^V}---\left(5\right),$

M wherein ^vfor the number of partitions of statistic histogram on the V passage, in this paper experiment, be made as 16;

be defined as and drop on pixel frequency in i subregion;

for the maximum magnitude of all frequencies, as color complexity

normalized factor; If the V passage of picture is constant,

be defined as and adopt following formula to calculate:

${\mathrm{\δ}}_{\max }^V=\sqrt{{(1-\frac{1}{m^V})}^2+\frac{m^V-1}{{\left(m^V\right)}^2}}---\left(6\right),$

Last color complexity is:

$S_C=\frac{S_C^H+S_C^S+S_C^V}{3}---\left(7\right);$

Step 12, this method adopts a cubic metre of earth displacement (Earth Mover ' s Distances, be called for short the EMD distance, the detailed description of this tolerance is published in the paper " The Earth Mover ' s Distance as a Metric for Image Retrieval " of International Journal ofComputerVision referring to YOSSI RUBNER etc.) calculate the similarity in twos between picture.EMD is apart from the alignment cost E (G with the statistic histogram of two width pictures under a certain feature _i, G _i', { f _ij) as the standard of weighing the picture analogies degree, this process is described as follows:

Minimize: $E(D_I,G_{I^{\′}},\{f_{\mathrm{ij}}\left\}\right)={\mathrm{\Σ}}_{i=1}^{24}{\mathrm{\Σ}}_{j=1}^{24}{f}_{\mathrm{ij}}{d}_{\mathrm{ij}},$

Meet prerequisite: f _ij>=0 i ∈ [1,24] wherein, j ∈ [1,24],

$\underset{j}{\mathrm{\Σ}}=f_{\mathrm{ij}}\≤g_i$ I ∈ [1,24] wherein,

$\underset{i}{\mathrm{\Σ}}=f_{\mathrm{ij}}\≤{g^{\′}}_i$ J ∈ [1,24] wherein,

$\underset{i=1}{\overset{24}{\mathrm{\Σ}}}\underset{j=1}{\overset{24}{\mathrm{\Σ}}}{f}_{\mathrm{ij}}=\min (\underset{i=1}{\overset{24}{\mathrm{\Σ}}}{g}_i,\underset{j=1}{\overset{24}{\mathrm{\Σ}}}{g^{\′}}_j),$

G wherein _i={ g _i, i=1 ..., 24}, G _i,={ g ' _j, j=1 ..., 24},

EMD is apart from the alignment cost with the statistic histogram of two width pictures under a feature as weighing the picture analogies degree, and EMD is apart from adopting following formula to calculate:

G _I＝{g _i，i＝1，…，24}，G _I，＝{g′ _j，j＝1，…，24} (8)，

$\mathrm{EMD}(G_I,G_{I^{\′}})=\frac{{\mathrm{\Σ}}_{i=1}^{24}{\mathrm{\Σ}}_{j=1}^{24}{f}_{\mathrm{ij}}{d}_{\mathrm{ij}}}{{\mathrm{\Σ}}_{i=1}^{24}{\mathrm{\Σ}}_{j=1}^{24}{f}_{\mathrm{ij}}}---\left(9\right);$

G in formula (8) _iand G _i, be respectively the hsv color spatial histogram proper vector of picture I and picture I ' correspondence, wherein the H passage is divided into 16 sub-blocks, and S passage and V passage all are divided into 4 sub-blocks, three common color feature vector G that form one 24 dimension of passages _iand G _i', g wherein _iand g ' _jrepresent the sub-block of dividing in subchannel.Formula (9) calculates the EMD distance that obtains picture I and picture I ', middle f _ijmean g _iand g ' _jstream between two sub-blocks (flow), d _ijmean g _iand g ' _jl1 distance between two sub-blocks.

This method is applied quick EMD computing method (this algorithm can be referring to Ofir Pele, and Michael Werman is published in the paper " Fast and Robust Earth Mover ' s Distances " of ICCV) and is obtained in pictures in each width picture and pictures EMD distance between other pictures and choose the minimum value wherein identification S as this picture _d.

Step 13, the importance degree S of last picture _ibe defined as:

S _I＝S _C+ωS _D （10）；

Wherein ω is for controlling weighing factor between the two, and the ω span is the real number between 0 ~ 1, and ω all gets 0.3 in the method.

In step 2, utilize in step 1 the importance degree information of the every width picture obtained to determine the circle that a series of radiuses are associated with importance degree information, then by circle, fill (Circle Packing) algorithm and realize that viewing area divides to determine the viewing area of picture in net result.The target of circle filling algorithm is the circle for a series of different radiis, and under the condition that allows the homogeneous convergent-divergent, the plane container of putting into a given shape that all circles are compacted, finally also can conveniently calculate and comprise each subregion division of circle separately according to result.

Step 2 specifically comprises the following steps:

Step 21, the initial position in the setting center of circle.Pictures for given display panel Ω and total n width picture, the display panel center is placed in to the two-dimentional right-handed coordinate system origin position of (coordinate system has X and two coordinate axis of Y), then generate at random the initial position of n point as the center of circle in the coordinate system range at display panel place, the center of circle is sat according to its X to target value is ascending to be sorted, the identical situation for the X coordinate figure, the little center of circle of Y coordinate figure is front, ascending to a label i of each center of circle distribution, the integer that the span of i is 1～n, center of circle i corresponding circle C _i, the initial radium of circle is R _i;

Step 22, set round initial radium R _i.Picture in pictures is corresponding with the circle institute of a label respectively, and its corresponding complexity is distance B in comparison step 11 between nearest two centers of circle of the rear acquisition of distance between any two, the center of circle _min, round initial radium R _icomputing formula is as follows:

$R_i=\left\{\begin{array}{cc}\frac{D_{\min }}{2},& i=1;\\ \frac{S_I^i*R_1}{S_I^1},& 1<i\&le;n;\end{array}\right.---\left(11\right);$

Wherein

for the complexity of the label picture that is 1,

for the complexity of the label picture that is i, R ₁the radius of a circle that label is 1;

Step 23, the circle fill method is justified filling to display panel Ω zone.After obtaining the initial home position and initial radium of circle, under the prerequisite of non-intersect and all circle in the zone of display panel Ω between assurance circle and circle, all circles are amplified in proportion, after having amplified, dynamically adjust home position, repeat the amplification adjustment process, until can't continue to amplify the time, stop this process, and obtain the arrangement of compacting in display panel Ω zone inner circle.

The circle Main Function of filling algorithm in step 2 be exactly in the Ω of given area, obtain a regional inner circle all nonoverlapping compacting " layout " (Conguration), wherein round filling algorithm specifically describes as follows:

For a given display panel

with n and the corresponding circle of particular picture in pictures

corresponding radius of a circle is

the circle filling algorithm is in step.When algorithm initialization, initialization zoom factor of given all circles

k=1, circle C _ibecome kC _i, the circle filling algorithm can be described as with next optimization problem:

Maximize: k

Meet prerequisite:

i ∈ 1 ..., n}

i，j∈{1，…，n}，i≠j.

The home position obtained X=(x for set ₁..., x _n) expression, wherein x _imean circle C _icentral coordinate of circle.This method is called one " layout " by X, if two constraints above-mentioned of all satisfactory foots, this claims that X is one " efficient layout ".

For above-mentioned optimizing process, this method adopts a kind of round filling algorithm based on weights figure (Power diagram) to be solved.

Weights figure is a kind of cum rights Voronoi diagram (Voronoi diagram, can participate in about the detailed introduction of Voronoi diagram the paper " Centroidalvoronoi tessellations:Applications andalgorithms " that Q Du etc. is published in SIAM)

for a different set of point, some x _ia weights ω is separately arranged _i>=0, i=1 ..., n.In given area, some x is to some x _iweights distances (Power distance) d _ω(x, x _i) be defined as follows:

d _ω(x，x _i)＝||x-x _i|| ²-ω _i （12）；

Then according to d _ω(x, x _i) divide definition and some x to carrying out zone _icorresponding regional Q (x _i) as follows:

All Q (x _i) a weights figure of the point set X that forms of set.

A given zone Ω _imean Q (x _i) with the common factor of Ω:

Ω _i＝Q(x _i)∩Ω （14）；

All Ω _ithe set formed has formed weights figure, wherein a Ω of a point set X in regional Ω _ibe called as and an x _icorresponding " cell " (Cell).

Be applied in this method, weights are ω _isome x _iexpression is with an x _ifor the center of circle,

the circle of radius, i.e. ω _i=R _i ².And because the prerequisite that optimizing process need to be satisfied can not be overlapping for circle and circle, so all circles all are contained in center of circle x _i" cell " Ω _iin, can not exceed cell Ω _iscope.

In order to calculate Voronoi diagram, this method adopts the method (the method specifically can be published in referring to S Lloyd etc. the paper " Least squares quantization in PCM " of IEEE Transactions on Information Theory) of Lloyd dynamically to adjust the position in the center of circle, until obtain maximum zoom factor k, obtain an arrangement of compacting under the prerequisite that guarantees the radius ratio between circle for the given number circle in given area according to zoom factor k, circle is filled and is so far stopped.

Step 24, the subregion of display panel is divided, and obtains the display space of every width picture.The final position of circle in display panel Ω obtained according to step 23, the perpendicular bisector of the circle center line connecting of two adjacent circles is set between any two adjacent circles, circumference encloses the subregion that closed polygon that all perpendicular bisectors are staggered to form has formed display panel, wherein comprises round C _ipolygon Q _pbe the corresponding display space of picture i associated with it.。

Step 3, in the display parameter optimizing process, set by calculating in the display space that makes every width picture obtain in step 2 towards angle, putting position and convergent-divergent yardstick of every width picture and present substantially its main contents.Step 3 specifically comprises the following steps:

Step 31, set picture towards angle.This method give one, every width picture random towards angle θ, θ meets [θ _m, θ _m] interior being evenly distributed of scope, wherein θ _mfor maximum allows deflection angle, this method arranges θ _mbe 30 °, thereby make picture presentation there is visual diversity, rather than dull inflexible.

Step 32, determine coordinate and scaled size.

At first, obtain picture importance degree figure.Method by propositions such as Cheng Mingming is calculated the importance degree figure (SaliencyMap) (this some algorithm can be published in referring to Cheng Mingming etc. the paper " Global Contrast based SalientRegionDetection " of CVPR) of picture, the importance degree that reflects each pixel by the importance value (Saliency) that in importance degree figure, each pixel is corresponding, importance value is larger, shows that the probability that this pixel can show in final splicing result is larger; In order to guarantee that the common more responsive people's face of people can fully be shown, use people's face detection algorithm in OpenCV (this algorithm can be published in referring to PaulViola and MichaelJones CVPR " RapidObjectDetectionusingaBoostedCascadeofSimpleFeature ") to detect human face region, the importance value of importance degree figure in this zone is set to maximum.

Then, obtain picture ROI zone.The importance degree figure obtained is carried out to binary conversion treatment based on threshold value, threshold value is the real number between 0 ~ 1, this method is set as 0.618, then (corrode with expansion and all belong to morphologic base conditioning means through corroding with dilation procedure, concrete grammar can be referring to JY Gil, R Kimmel is published in paper " the Efficient dilation of Pattern Analysis andMachine Intelligence, erosion, opening, and closing algorithms ") obtain several connected regions, after sorting from large to small according to the connected region area, pass through Sklansky, J. (this Part Methods can be referring to Sklansky for the Minimum Convex Closure algorithm proposed, J. be published in the paper " Finding the Convex Hull of a Simple Polygon " on Pattern Recognition Letters) with convex polygon surround area sort first three/mono-simply connected region, gained polygon Q _rbe picture ROI zone.

Finally, determine picture coordinate and convergent-divergent yardstick.Q _rand Q _pmean respectively picture ROI zone polygon and the corresponding displaying drawing board of picture display space.At first, by Q _rcenter of gravity P _rin Q _pcenter, determine before the θ of angle, convergent-divergent yardstick when this picture covers spacial flex just fully is the initial value of convergent-divergent yardstick.With P _rfor initial point builds the cartesian coordinate system O that is parallel to the drawing board coordinate system.Then, in definition O, the information dropout value S in all quadrants is at Q _rin but not at Q _pthe mean value of the importance value of interior pixel.S _tl, S _tr, S _brand S _b1mean respectively upper left, upper right, the information dropout value in four quadrants in bottom right and lower-left.A didactic moving direction vector v is determined by following formula:

v＝((S _tl+S _b1)-(S _tr+S _br)，(S _tl+S _tr)-(S _bl+S _br)) （15）；

V is normalized to vector of unit length

be next step moving direction on display panel of picture.The maximum movement with P _rcentered by construct Gaussian distribution:

$N(\mathrm{Pr}+\frac{r}{2}\&CenterDot;\hat{v},r)---\left(16\right);$

The maximum moving distance that wherein r is each picture.Distribute and get 10 position candidate accordingly, and picture is moved to the minimum position candidate of information dropout value S.On new position, will determine new convergent-divergent yardstick at this according to the method for determining convergent-divergent yardstick initial value.

The mobile convergent-divergent step of above search iteration is carried out several times, after convergence or surpassing a certain predetermined maximum iteration time restriction (this method is set to 30 times), stops iteration, information loss value minimum now, and picture coordinate and convergent-divergent yardstick have been determined.Searching algorithm has guaranteed that picture can show its main information on limited distribution gained subregion.

Step 4 specifically comprises the following steps:

To each pixel on display panel, equal associated one group of probability { Prob ₁(p), Prob ₂(p) ..., Prob _n(p) }, wherein n is the picture sum, Prob _i(p) mean the probability that i width picture can show in this pixel place respective pixel.Q _piand Q _ricorresponding display panel spacial flex and its ROI zone of difference i width picture.Mixing can be so that Q _riin not at Q _piin part also can obtain representing to a certain extent, Prob _i(p) computing formula is as follows:

${\mathrm{Prob}}_i\left(p\right)=\left\{\begin{array}{cc}1& p\&Element;Q_{\mathrm{pi}},p\&Element;Q_{\mathrm{ri}}\\ e^{\frac{-d(p,Q_{\mathrm{ri}})}{2\mathrm{\&sigma;}}}& p\&Element;Q_{\mathrm{pi}},p\&NotElement;Q_{\mathrm{ri}}\\ e^{\frac{-d(p,Q_{\mathrm{pi}})}{2\mathrm{\&sigma;}}}& p\&NotElement;Q_{\mathrm{pi}},p\&Element;Q_{\mathrm{ri}}\\ 0& p\&NotElement;Q_{\mathrm{pi}},p\&NotElement;Q_{\mathrm{ri}}\end{array}\right.---\left(17\right);$

Wherein, the end that e is natural logarithm,, value is 2.71828 ..., d (p, Q _pi) mean that some p is to display space, Q _pinearest Euclidean distance, d (p, Q _ri) mean that some p is to ROI zone Q _rinearest Euclidean distance, the standard deviation that σ is contribution rate probability probability distribution, calculate σ ≈ 0.39894223.

After the association probability of each pixel of display panel calculates by above rule, average being distributed on its 4 neighborhood again, so that contiguous pixel has approaching probable value again by one group of association probability normalization of each pixel, the probability graph obtained thereafter as the Alpha channel value of married operation for finally piecing the synthetic of picture together.

Beneficial effect: the present invention includes following advantage:

(1) viewing area is supported more flexibly.Support the result of piecing together of boundary with any contour, be supported in simultaneously and piece the scaling that carries out sub-pictures on result, rotation and the personalized customization operation such as switch in twos together.

(2) higher extensibility has both low coupling simultaneously.This characteristic has benefited from following three aspects: at first, the efficient robust of subregion partitioning algorithm that this paper adopts can be processed the subregion of 30 above pictures and divide in average 1 second.Secondly, the state parameter of every pictures---position, a little less than the optimization cross correlation of angle, convergent-divergent yardstick and level, each parameter all can be at a stage Optimization Solution independently.Finally, the state parameter optimization of every pictures is all closed in the state decoupling of contiguous picture, thereby can carry out concurrently.

(3) robustness of processing speed and Geng Gao faster.Have benefited from low coupling in (2) and the high efficiency of region partitioning algorithm, the speed of this method is not slower than the existing methods prestissimo, faster than the existing method of major part.Simultaneously, than the AutoCollage of Microsoft, this method can be done the pictures splicing of any amount, and AutoCollage to require the picture number of pictures minimum be 7 pictures.

(4) more can meet the objective demand of people to personage's photo.For the picture that the people is arranged, the present invention can detect to arrange by people's face and show that people's face is limit priority, thereby the situation of effectively having avoided people's face to be blocked by the viewing area of other photos, this has also been avoided the people's face occurred in the AutoCollage of Microsoft to show infull situation to a great extent; For nobody picture such as landscape, object, animal, the present invention can guarantee that the prospect part can be shown with larger probability, and the prospect part can be largely the main region of photo content, thereby obtains the good result that shows.

(5) better display effect.The present invention is in the process of picture discharge, picture towards the random selection by the certain angle scope towards, therefore the final more existing method of demonstration result is more natural, and the reasonable setting of angular range simultaneously can avoid whole result to give mixed and disorderly sensation.

The accompanying drawing explanation

Below in conjunction with the drawings and specific embodiments, the present invention is done further and illustrates, above-mentioned or otherwise advantage of the present invention will become apparent.

The process flow diagram that Fig. 1 is the inventive method.

Fig. 2 is that in the present invention, circle filling (Circle Packing) algorithm is divided the process in 5 zones in a rectangular area.

Fig. 3 is the workflow schematic diagram that in the present invention, viewing area is optimized.

Fig. 4 is the example flow chart that the inventive method is implemented.

The feedback result figure of Fig. 5 user investigation.

The embodiment that Fig. 6 is the animal picture collection finally generates result.

The embodiment that Fig. 7 is animation placard pictures finally generates result.

The embodiment that Fig. 8 is child's pictures finally generates result.

The embodiment that Fig. 9 is the flowers pictures finally generates result.

The embodiment that Figure 10 is the rhombus display panel finally generates result.

The embodiment that Figure 11 is oval display panel finally generates result.

The AutoCollage that Figure 12 a is football player's pictures finally generates result.

The present embodiment that Figure 12 b is football player's pictures finally generates result.

The AutoCollage that Figure 13 a is the doll pictures finally generates result.

The present embodiment that Figure 13 b is the doll pictures finally generates result.

Embodiment:

At first the flow process of this method assesses the significance level of each width picture as shown in Figure 1, then inferior importance value is mapped as to the input parameter of round filling algorithm---the initial radium value of circle; Then obtain regional division result according to the circle filling algorithm, 5 circles having showed a known initial radium ratio in Fig. 2, progressively increase the zoom factor of circle, obtain the maximal value k of a zoom factor after repeatedly calculating, then according to the circumscribed polygon of circle, obtain the zone division of display panel; Use a kind of heuristic search algorithm iteration to search for to optimize the display parameter of each width picture in assigned display space, Fig. 3 to show and once search for optimizing process; Be finally the overlapping algorithm between same seamless blended technical finesse picture, finally obtain pictures and piece the displaying result together.Fig. 4 has showed that pictures that have 5 width pictures shear the sub-result after the idiographic flow of splicing and every operation complete.

Specifically, as shown in Figure 1, the invention discloses image cut joining method in a kind of computing machine:

Step 1, importance degree assessment: weigh the color characteristic of every width picture in pictures, obtain the color complexity S of every width picture by the diversity of color in the statistics picture _c; Adopt a cubic metre of earth displacement method to calculate picture and concentrate the EMD distance between each picture, the similarity S using the EMD of every width picture and other pictures apart from minimum value other pictures in this picture and pictures _d, by S _cand S _dweighted sum as the importance degree S of this picture _i;

Step 2, the display panel subregion is divided the display space that obtains every width picture: choose the one group circle identical with picture number in pictures, wherein the ratio between the importance degree of ratio and the described every width picture between initial radium of each circle is identical, fill display panel by the circle fill method, divide and obtain the display panel subregion by the circumscribed polygon of circle after filling completes, thereby determine the display space Q of every width picture on display panel _p;

Step 3, display parameter optimization: determine every width picture towards angle θ, putting position, convergent-divergent yardstick; Calculate the importance degree figure of each picture, will contain in the picture of facial image the importance value of each pixel in the facial image zone and be set to maximal value; Pixel higher than default importance degree threshold value in each picture is formed to connected region, after sorting from large to small according to the connected region area, the connected region of area sequence front 1/3 is divided in a polygonal region, and this polygonal region is the ROI zone Q of this picture _r;

Step 4, seamless blended is played up, and the Based on Probability mixture model carries out seamless blended to the borderline region between the display panel subregion to be played up, thereby complete the computing machine picture, concentrates picture to shear splicing.

In step 1, the complexity by picture and pass judgment on the importance of picture with the similarity of other pictures.

Step 1 specifically comprises the following steps:

Step 11, the complexity of the statistics with histogram information definition color on use hsv color space, the color complexity on the H passage

adopt following formula to calculate:

$S_c^H=1-\frac{\sqrt{{\mathrm{\&Sigma;}}_{i=1}^{m^H}{(h_i^H-\frac{1}{m^H})}^2}}{{\mathrm{\&delta;}}_{\max }^H}---\left(1\right);$

M wherein ^hrepresent the number of partitions of statistic histogram on the H passage, in this paper experiment, be made as 16;

be defined as and drop on pixel frequency in i subregion;

for the maximum magnitude of all frequencies, as color complexity S _cnormalized factor.If the H passage of picture is constant, the histogram of H passage only has a subregion that value is arranged, at this moment

be defined as

${\mathrm{\&delta;}}_{\max }^H=\sqrt{{(1-\frac{1}{m^H})}^2+\frac{m^H-1}{{\left(m^H\right)}^2}}---\left(2\right);$

Complexity value on the S passage adopt following formula to calculate:

$S_c^S=1-\frac{\sqrt{{\mathrm{\&Sigma;}}_{i=1}^{m^S}{(h_i^S-\frac{1}{m^S})}^2}}{{\mathrm{\&delta;}}_{\max }^S}---\left(3\right);$

M wherein ^sfor the number of partitions of statistic histogram on the S passage, in this paper experiment, be made as 16;

be defined as and drop on pixel frequency in i subregion;

for the maximum magnitude of all frequencies, as color complexity

normalized factor; If the S passage of picture is constant,

be defined as and adopt following formula to calculate:

${\mathrm{\&delta;}}_{\max }^S=\sqrt{{(1-\frac{1}{m^S})}^2+\frac{m^S-1}{{\left(m^S\right)}^2}}---\left(4\right);$

Complexity value on the V passage

adopt following formula to calculate:

$S_c^V=1-\frac{\sqrt{{\mathrm{\&Sigma;}}_{i=1}^{m^V}{(h_i^V-\frac{1}{m^V})}^2}}{{\mathrm{\&delta;}}_{\max }^V}---\left(5\right),$

M wherein ^vfor the number of partitions of statistic histogram on the V passage, in this paper experiment, be made as 16;

be defined as and drop on pixel frequency in i subregion;

for the maximum magnitude of all frequencies, as color complexity

normalized factor; If the V passage of picture is constant, be defined as and adopt following formula to calculate:

${\mathrm{\&delta;}}_{\max }^V=\sqrt{{(1-\frac{1}{m^V})}^2+\frac{m^V-1}{{\left(m^V\right)}^2}}---\left(6\right),$

Last color complexity is:

$S_C=\frac{S_C^H+S_C^S+S_C^V}{3}---\left(7\right);$

Step 12, EMD is apart from the alignment cost with the statistic histogram of two width pictures under a feature as weighing the picture analogies degree, and EMD is apart from adopting following formula to calculate:

H _I＝{h _i，i＝1，…，24}，H _I，＝{h′ _j,j＝1，…，24} (8)。

$\mathrm{EMD}(H_I,H_{I^{\&prime;}})=\frac{{\mathrm{\&Sigma;}}_{i=1}^{24}{\mathrm{\&Sigma;}}_{j=1}^{24}{f}_{\mathrm{ij}}{d}_{\mathrm{ij}}}{{\mathrm{\&Sigma;}}_{i=1}^{24}{\mathrm{\&Sigma;}}_{j=1}^{24}{f}_{\mathrm{ij}}}---\left(9\right).$

H in formula (8) _iand H _i, be respectively the hsv color spatial histogram proper vector of picture I and picture I ' correspondence, wherein the H passage is divided into 16 sub-blocks, and S passage and V passage all are divided into 4 sub-blocks, three common color feature vector H that form one 24 dimension of passages _iand H _i', h wherein _iand h ' _jrepresentative means the sub-block of dividing in subchannel, and i is sub-block h _idimension label and j be sub-block h ' _jin the dimension label; Formula (9) calculates the EMD distance that obtains picture I and picture I ', wherein f _ijmean h _iand h ' _jstream between two sub-blocks, d _ijmean h _iand h ' _jl1 distance between two sub-blocks.

According to obtaining EMD distance between picture in twos, therefrom choose minimum value as the similarity S between other pictures in a picture and pictures _d;

Step 13, the importance degree of last picture is defined as:

S _I＝S _C+ωS _D (10)

ω similarity S _dfor controlling weighing factor between the two, ω value 0.3.

In step 2, utilize the importance degree information of the every width picture obtained in step 1 to determine the circle that a series of initial radiums are associated with importance degree information, then by the circle filling algorithm, realize the viewing area division.Specifically comprise the following steps:

Step 21, the initial position in the setting center of circle.Pictures for given display panel Ω and total n width picture, the display panel center is placed in to the two-dimentional right-handed coordinate system origin position of (coordinate system has X and two coordinate axis of Y), then generate at random the initial position of n point as the center of circle in the coordinate system range at display panel place, the center of circle is sat according to its X to target value is ascending to be sorted, the identical situation for the X coordinate figure, the little center of circle of Y coordinate figure is front, ascending to numbering i of each center of circle distribution, the integer that the span of i is 1～n, center of circle i corresponding circle C _i, the initial radium R of circle _imean.

Step 22, set round initial radium.Picture in pictures is corresponding with the circle institute of a label respectively, and its corresponding complexity is

as shown in Figure 4, picture 1 and C ₁correspondence, picture 2 and C ₂correspondence, picture 3 and C ₃correspondence, picture 4 and C ₄correspondence, picture 5 and C ₅corresponding; Distance B in comparison step 11 between nearest two centers of circle of the rear acquisition of distance between any two, the center of circle _min, round initial radium R _icomputing formula is as follows:

$R_i=\left\{\begin{array}{cc}\frac{D_{\min }}{2},& i=1;\\ \frac{S_I^i*R_1}{S_I^1},& 1<i\&le;n;\end{array}\right.---\left(11\right);$

Wherein

for the complexity of the label picture that is 1,

complexity for the label picture that is i; As shown in the subregion of " circle that radius is relevant to corresponding picture importance degree " sign in Fig. 4: obtaining the corresponding radius of a circle initial value of picture 1 after calculating is 0.762, the corresponding radius of a circle initial value of picture 2 is 0.801, the corresponding radius of a circle initial value of picture 3 is 0.823, the corresponding radius of a circle initial value of picture 4 is 0.824, and the corresponding radius of a circle initial value of picture 5 is 0.713;

Step 23, the circle fill method is justified filling to the display panel zone.After obtaining as shown in Figure 2 the initial home position and initial radium of circle, under prerequisite in the zone of non-intersect and all circle at display panel between assurance circle and circle, all circles are amplified in proportion, after having amplified, dynamically adjust home position, repeat the amplification adjustment process, until can't continue to amplify the time, stop this process, obtain the arrangement of compacting in display panel zone inner circle, now justifying of correspondence unified zoom factor k=186.625, circle C ₁~ circle C ₅final position as shown in Figure 2.

Step 24, the subregion of display panel is divided, and obtains the display space of every width picture.The rank results of compacting of the viewing area inner circle obtained according to step 23, circle is adjacent the circumscribed polygon that common tangent between circle can form this circle, for nontangential situation between circle and adjacency circle, this method is chosen with two perpendicular bisectors of round line and is participated in polygonal formation.The polygon obtained has formed the subregion of display panel to be divided, and as shown in dotted line in Fig. 2 is divided, wherein comprises round C _ipolygon Q _pbe the corresponding display space of picture associated with it.

In step 3, determine as shown in Figure 3 picture towards angle, the display parameter such as coordinate and convergent-divergent yardstick, make maximized its important content that presents in the viewing area of the limited size of display panel that every width picture can be corresponding at it.

As shown in Figure 3, step 3 specifically comprises the following steps:

Step 31, set picture towards angle.This method give one, every width picture random towards angle θ, θ meets [θ _m, θ _m] interior being evenly distributed of scope, wherein θ _mfor maximum allows deflection angle, this method arranges θ _mbe 30 °, as shown in Figure 3, that sets picture in Fig. 3 in " set picture towards " step is set to 30 ° towards angle θ.

Step 32, determine coordinate and scaled size.

At first, obtain picture importance degree figure.As shown in " calculating of picture importance degree " step in Fig. 4, calculate the importance degree figure of picture by the method for the propositions such as Cheng Mingming, the corresponding importance degree figure of the picture obtained is as shown in " importance degree figure " subregion in Fig. 4, then use the people's face detection algorithm in OpenCV to detect human face region, the importance value of importance degree figure in this zone is set to maximum.

Then, obtain picture ROI zone.The importance degree figure obtained is carried out to binary conversion treatment based on threshold value 0.618, then obtain several connected regions through corroding with dilation procedure, after sorting from large to small according to the connected region area, as shown in Fig. 4, " calculating ROI zone approximate polygon ", pass through Sklansky, J. the Minimum Convex Closure algorithm proposed with a convex polygon encirclement area sort first three/mono-simply connected region, gained polygon Q _rbe picture ROI zone, polygon is as shown in " ROI zone " subregion in Fig. 4.

Finally, determine picture coordinate and convergent-divergent yardstick.Q as shown in Figure 3 _rand Q _pmean respectively picture ROI zone polygon and the corresponding displaying drawing board of picture display space.At first, by Q _rcenter of gravity P _rin Q _pcenter, determine before the θ of angle, convergent-divergent yardstick when this picture covers spacial flex just fully is the initial value of convergent-divergent yardstick.With P _rfor initial point builds the cartesian coordinate system O that is parallel to the drawing board coordinate system.Then, in definition O, the information dropout value S in all quadrants is at Q _rin but not at Q _pthe mean value of the importance value of interior pixel.S _tl, S _tr, S _brand S _b1mean respectively upper left, upper right, the information dropout value in four quadrants in bottom right and lower-left.A didactic moving direction vector v is determined by following formula:

v＝((S _tl+S _b1)-(S _tr+S _br)，(S _tl+S _tr)-(S _bl+S _br)) （12）；

V is normalized to vector of unit length

be next step moving direction on display panel of picture.With P _rcentered by construct Gaussian distribution:

$N(\mathrm{Pr}+\frac{r}{2}\&CenterDot;\hat{v},r)---\left(13\right);$

The maximum moving distance that wherein r is each picture.Distribute and get 10 position candidate accordingly, and picture is moved to the minimum position candidate of information dropout value S.On new position, will determine new convergent-divergent yardstick at this according to the method for determining convergent-divergent yardstick initial value.

The mobile convergent-divergent step of above search iteration is carried out several times, after convergence or surpassing a certain predetermined maximum iteration time restriction (this method is set to 30 times), stops iteration, and picture coordinate and convergent-divergent yardstick have been determined.

In step 4, as shown in Fig. 4 " seamless blended based on importance degree figure information is played up " step, the importance degree information of picture, in conjunction with entering mixed process, is obtained to the final splicing result of shearing.

To each pixel on display panel, equal associated one group of probability { Prob ₁(p), Prob ₂(p) ..., Prob _n(p) }, wherein n is the picture sum, Prob _i(p) mean the probability that i width picture can show in this pixel place respective pixel.Q _piand Q _ricorresponding display panel spacial flex and its ROI zone of difference i width picture.Mixing can be so that Q _riin not at Q _piin part also can obtain representing to a certain extent, Prob _i(p) computing formula is as follows:

${\mathrm{Prob}}_i\left(p\right)=\left\{\begin{array}{cc}1& p\&Element;Q_{\mathrm{pi}},p\&Element;Q_{\mathrm{ri}}\\ e^{\frac{-d(p,Q_{\mathrm{ri}})}{2\mathrm{\&sigma;}}}& p\&Element;Q_{\mathrm{pi}},p\&NotElement;Q_{\mathrm{ri}}\\ e^{\frac{-d(p,Q_{\mathrm{pi}})}{2\mathrm{\&sigma;}}}& p\&NotElement;Q_{\mathrm{pi}},p\&Element;Q_{\mathrm{ri}}\\ 0& p\&NotElement;Q_{\mathrm{pi}},p\&NotElement;Q_{\mathrm{ri}}\end{array}\right.---\left(14\right).$

Embodiment

The present embodiment for the hardware environment of test is: Intel-Core2Duo4.2GHz processor, 4G internal memory.Software environment is

visual Studio2010 and

the Windows7 professional version.Test pattern comes from disclosed animated film placard and the photos such as other some animals and flowers on the interior photo of group, network.Being divided into 9 groups according to macrotaxonomy during experiment, is respectively animal, doll, football player, wineglass, animation placard, cat and dog, child, flowers, grey clothing schoolgirl.

The present embodiment is according to the difference of the picture number of input pictures, and the time loss of piecing together did not wait by tens seconds at several seconds, and consume at people's face and detect and importance degree figure calculating section main computing time.Simultaneously also by the experimental result of this method with the result that " automatically piecing together " function (AutoCollage) in Photo Gallery external member obtains has been carried out user investigation.The classmate of 124 bit machines systems has participated in the user investigation of double blinding, is not having under other extraneous prerequisites of intervening judgement on one's own account select more to meet own aesthetic result.Fig. 5 has shown the feedback result of user investigation, investigation result has been carried out normalized, the preference of recently showing the user with percentage, left side black cylinder means that the user who selects the present embodiment to generate result accounts for the number percent of total number of persons, and right side white cylinder means that the user who selects AutoCollage to generate result accounts for the number percent of total number of persons.Specifically referring to, animal (embodiment of Fig. 6 the present embodiment animal picture collection finally generates result schematic diagram), (AutoCollage that Figure 13 a is the doll pictures finally generates result schematic diagram to doll, the present embodiment that Figure 13 b is the doll pictures finally generates result schematic diagram), (AutoCollage that Figure 12 a is football player's pictures finally generates result schematic diagram to the football player, the present embodiment that Figure 12 b is football player's pictures finally generates result schematic diagram), wineglass, (embodiment of Fig. 7 the present embodiment animation placard pictures finally generates result schematic diagram to the animation placard, cat and dog, child's (embodiment that Fig. 8 is child's pictures finally generates result schematic diagram), flowers (embodiment that Fig. 9 is the flowers pictures finally generates result schematic diagram), ash clothing schoolgirl, corresponding as a result than being followed successively by: 61.3%:38.7%, 66.1%:33.9%, 80.8%:19.2%, 69.6%:30.4%, 77.8%:22.2%, 59.5%:40.5%, 69.8%:30.2%, 75.4%:24.6%, 58.0%:42.0%.

Fig. 5 is that user investigation user investigation result as a result shows, the result that all 9 groups of results generate for AutoCollage has advantage in various degree, and the 3rd group of football player's result has relatively reached 80.8% to 19.2%.Two kinds of final pictures that obtain of method are compared as Figure 12 a, Figure 12 b, and shown in Figure 13 a, Figure 13 b.The result of user investigation has also shown that most of user tends to the pictures splicing result of selecting this method to generate.On user's questionnaire, also be provided with " selection reason " this choosing and answer item, from answering the consumers' opinions of this problem, the reason of most of selection this method is all to concentrate on following two:

1, the picture arrangement that this method generates is in picturesque disorder, generates the inflexible arrangement of result with respect to AutoCollage more natural, random, makes us pleasing.This point relatively can be found out from Figure 12 a and Figure 12 b and Figure 13 a and Figure 13 b's, the result of AutoCollage arrange be the subregion of every a line substantially on the same horizontal line, and arranging of this method obtained zone more flexibly and arranged because the circle fill method carries out zone division;

2, the situation that in the picture that this method generates, people's face or object block all is better than the generation result of AutoCollage.The blocking contrast and can be embodied in the example of Figure 12 a and Figure 12 b of people's face;

Figure 10 and Figure 11 are when given display panel shape is respectively rhombus and ellipse, the net result that the present embodiment obtains, and AutoCollage only supports three kinds of rectangle display panels under size.

The pictures that during due to user investigation, can't allow the user scene experience this method and AutoCollage are pieced formation speed together, and the processing speed advantage of this method and other only also just can not obtain embodiment by user investigation by finally piecing the advantage that result can't obtain together.

Claims (5)

1. a computing machine picture concentrates picture to shear joining method, it is characterized in that, comprises the following steps:

Step 1, importance degree assessment: calculate the color characteristic that picture is concentrated every width picture, obtain the color complexity S of every width picture by the diversity of color in the statistics picture _c; Adopt a cubic metre of earth displacement method to calculate picture and concentrate the EMD distance between each picture, the similarity S using the EMD of every width picture and other pictures apart from minimum value other pictures in this picture and pictures _d, by S _cand S _dweighted sum as the importance degree S of this picture _i;

Step 2, the display panel subregion is divided the display space that obtains every width picture: choose the one group circle identical with picture number in pictures, wherein the ratio between the importance degree of ratio and the described every width picture between initial radium of each circle is identical, fill display panel by the circle fill method, divide and obtain the display panel subregion by the circumscribed polygon of circle after filling completes, thereby determine the display space Q of every width picture on display panel _p;

Step 3, display parameter optimization: determine every width picture towards angle θ, putting position, convergent-divergent yardstick; At first in given range, set each width picture towards angle θ; Calculate the importance degree figure of each picture, will contain in the picture of facial image the importance value of each pixel in the facial image zone and be set to maximal value; Pixel higher than default importance degree threshold value in each picture is formed to connected region, after sorting from large to small according to the connected region area, the connected region of area sequence front 1/3 is divided in a polygonal region, and this polygonal region is the ROI zone Q of this picture _r; Then the zone of the ROI towards angle θ, the picture Q set according to picture _rand definite display space Q corresponding to picture in step 2 _p, determine picture putting position and convergent-divergent yardstick;

Step 4, seamless blended is played up, and the Based on Probability mixture model carries out seamless blended to the borderline region between the display panel subregion to be played up, thereby complete the computing machine picture, concentrates picture to shear splicing.

2. in a kind of computing machine according to claim 1, picture is sheared joining method, it is characterized in that, described step 1 specifically comprises the following steps:

Step 11, the complexity of the statistics with histogram information definition color on use hsv color space, the color complexity on the H passage

adopt following formula to calculate:

$S_c^H=1-\frac{\sqrt{{\mathrm{\&Sigma;}}_{i=1}^{m^H}{(h_i^H-\frac{1}{m^H})}^2}}{{\mathrm{\&delta;}}_{\max }^H}---\left(1\right),$

M wherein ^hthe number of partitions for statistic histogram on the H passage; for dropping on pixel frequency in i subregion;

for the maximum magnitude of all frequencies, as color complexity

normalized factor; If the H passage of picture is constant, the histogram of H passage only has a subregion that value is arranged,

be defined as and adopt following formula to calculate:

${\mathrm{\&delta;}}_{\max }^H=\sqrt{{(1-\frac{1}{m^H})}^2+\frac{m^H-1}{{\left(m^H\right)}^2}}---\left(2\right),$

Complexity value on the S passage

adopt following formula to calculate:

$S_c^S=1-\frac{\sqrt{{\mathrm{\&Sigma;}}_{i=1}^{m^S}{(h_i^S-\frac{1}{m^S})}^2}}{{\mathrm{\&delta;}}_{\max }^S}---\left(3\right),$

M wherein ^sthe number of partitions for statistic histogram on the S passage;

for dropping on pixel frequency in i subregion;

for the maximum magnitude of all frequencies, as color complexity normalized factor; If the S passage of picture is constant, be defined as and adopt following formula to calculate:

${\mathrm{\&delta;}}_{\max }^S=\sqrt{{(1-\frac{1}{m^S})}^2+\frac{m^S-1}{{\left(m^S\right)}^2}}---\left(4\right),$

Complexity value on the V passage adopt following formula to calculate:

$S_c^V=1-\frac{\sqrt{{\mathrm{\&Sigma;}}_{i=1}^{m^V}{(h_i^V-\frac{1}{m^V})}^2}}{{\mathrm{\&delta;}}_{\max }^V}---\left(5\right),$

M wherein ^vthe number of partitions for statistic histogram on the V passage;

for dropping on pixel frequency in i subregion;

for the maximum magnitude of all frequencies, as color complexity normalized factor; If the V passage of picture is constant,

be defined as and adopt following formula to calculate:

${\mathrm{\&delta;}}_{\max }^V=\sqrt{{(1-\frac{1}{m^V})}^2+\frac{m^V-1}{{\left(m^V\right)}^2}}---\left(6\right),$

Final color complexity is:

$S_C=\frac{S_C^H+S_C^S+S_C^V}{3}---\left(7\right);$

Step 12, EMD is apart from the alignment cost with the statistic histogram of two width pictures under a feature as weighing the picture analogies degree, and EMD is apart from adopting following formula to calculate:

G _I＝{g _i，i＝1，…，24}，G _I，＝{g′ _j，j＝1，…，24}(8)，

$\mathrm{EMD}(G_I,G_{I^{\&prime;}})=\frac{{\mathrm{\&Sigma;}}_{i=1}^{24}{\mathrm{\&Sigma;}}_{j=1}^{24}{f}_{\mathrm{ij}}{d}_{\mathrm{ij}}}{{\mathrm{\&Sigma;}}_{i=1}^{24}{\mathrm{\&Sigma;}}_{j=1}^{24}{f}_{\mathrm{ij}}}---\left(9\right),$

G in formula (8) _iand G _i, be respectively the hsv color spatial histogram proper vector of picture I and picture I ' correspondence, wherein the H passage is divided into 16 sub-blocks, and S passage and V passage all are divided into 4 sub-blocks, three common color feature vector G that form one 24 dimension of passages _iand G _i', g wherein _iand g ' _jrepresentative means the sub-block of dividing in subchannel, and i is sub-block g _idimension label and j be sub-block g ' _jin the dimension label; Formula (9) calculates the EMD distance that obtains picture I and picture I ', wherein f _ijmean g _iand g ' _jstream between two sub-blocks, d _ijmean g _iand g ' _jl1 distance between two sub-blocks;

According to EMD distance between picture in twos, therefrom choose minimum value as the similarity S between other pictures in a picture and pictures _d;

Step 13, the importance degree of last picture is defined as:

S _I＝S _C+ωS _D(10)，

ω similarity S _dfor controlling weighing factor between the two, the ω span is the real number between 0 ~ 1.

3. a kind of computing machine picture according to claim 2 concentrates picture to shear joining method, it is characterized in that, in step 2, comprises the following steps:

Step 21, set the initial position in the center of circle, for given display panel Ω and the pictures that comprise n width picture, the display panel center is placed in to the origin position of a two-dimentional right-handed coordinate system, coordinate system has X, two coordinate axis of Y, the random initial position of n point as the center of circle that generate in the coordinate system range at display panel Ω place, the center of circle is sat according to its X to target value is ascending to be sorted, the identical situation for the X coordinate figure, the little center of circle of Y coordinate figure is front, ascending to numbering i of each center of circle distribution, the integer that the span of i is 1～n, center of circle i corresponding circle C _i,

Step 22, set round initial radium R _i, the picture in pictures is corresponding with the circle institute of a label respectively, and the complexity that picture i is corresponding is

distance B in comparison step 21 between nearest two centers of circle of the rear acquisition of distance between any two, the center of circle _min, the initial radium R that label is the i circle _icomputing formula is as follows:

$R_i=\left\{\begin{array}{cc}\frac{D_{\min }}{2},& i=1;\\ \frac{S_I^i*R_1}{S_I^1},& 1<i\&le;n;\end{array}\right.---\left(11\right),$

Wherein

for the complexity of the label picture that is 1,

for the complexity of the label picture that is i, R ₁for the label radius of a circle that is 1;

Step 23, adopt the circle fill method to be justified filling to display panel Ω zone, after obtaining the initial home position and initial radium of circle, all circles are amplified in proportion, constraint condition for circle with round between non-intersect and all circle in the zone of display panel, after having amplified, adjust home position, repeat the amplification adjustment process, until can't continue to amplify the time, stop this process, obtain the final position of circle in display panel Ω zone;

Step 24, divide the subregion of display panel Ω, obtain the display space of every width picture: the final position of circle in display panel Ω obtained according to step 23, the perpendicular bisector of the circle center line connecting of two adjacent circles is set between any two adjacent circles, the closed polygon that all perpendicular bisectors are staggered to form has formed the subregion of display panel, wherein comprises round C _ipolygon Q _pbe the corresponding display space of picture i associated with it.

4. a kind of computing machine picture according to claim 3 concentrates picture to shear joining method, it is characterized in that, step 3 comprises the following steps:

Step 31, set picture towards angle: for every width picture arrange one random towards angle θ, θ span [θ _m, θ _m], θ wherein _mfor maximum allows deflection angle;

Step 32, determine coordinate and convergent-divergent yardstick: calculate the importance degree figure of each picture, will contain in the picture of facial image the importance value of each pixel in the facial image zone and be set to maximal value; Pixel higher than default importance degree threshold value in each picture is formed to connected region through expansion and erosion operations, after sorting from large to small according to the connected region area, the simply connected region of area sequence front 1/3 is divided in a Convex Polygon Domain, and this polygonal region is the ROI zone Q of this picture _r; Alternately iteration is found the optimized coordinate of every width picture and zoom factor.

5. a kind of computing machine picture according to claim 4 concentrates picture to shear joining method, it is characterized in that, in step 4, comprises the steps:

To each the pixel p on display panel, associated one group of contribution rate { Prob ₁(p), Prob ₂(p) ...., Prob _n(p) }, wherein n is the picture sum, Prob _i(p) mean the contribution rate of i width picture at this pixel place, 1≤i≤n; If Q _piand Q _ridifference corresponding display space and the ROI zone of i width picture; Prob _i(p) adopt following formula to calculate:

${\mathrm{Prob}}_i\left(p\right)=\left\{\begin{array}{cc}1& p\&Element;Q_{\mathrm{pi}},p\&Element;Q_{\mathrm{ri}}\\ e^{\frac{-d(p,Q_{\mathrm{ri}})}{2\mathrm{\&sigma;}}}& p\&Element;Q_{\mathrm{pi}},p\&NotElement;Q_{\mathrm{ri}}\\ e^{\frac{-d(p,Q_{\mathrm{pi}})}{2\mathrm{\&sigma;}}}& p\&NotElement;Q_{\mathrm{pi}},p\&Element;Q_{\mathrm{ri}}\\ 0& p\&NotElement;Q_{\mathrm{pi}},p\&NotElement;Q_{\mathrm{ri}}\end{array}\right.---\left(12\right),$

Wherein, the end that e is natural logarithm,, value is 2.71828 ..., d (p, Q _pi) mean that some p is to display space Q _pinearest Euclidean distance, d (p, Q _ri) mean the nearest Euclidean distance of some p to the regional Qri of ROI, the standard deviation that σ is contribution rate probability probability distribution;

Contribution rate Prob by each pixel _i(p) average being distributed on its 4 neighborhood, then by one group of contribution rate normalization of each pixel, the Alpha channel value that probability graph obtained is played up as seamless blended.

Images