CN104952065A - Method for building multilayer detailed skeleton model of garment images - Google Patents

Method for building multilayer detailed skeleton model of garment images Download PDF

Info

Publication number
CN104952065A
CN104952065A CN201510237236.1A CN201510237236A CN104952065A CN 104952065 A CN104952065 A CN 104952065A CN 201510237236 A CN201510237236 A CN 201510237236A CN 104952065 A CN104952065 A CN 104952065A
Authority
CN
China
Prior art keywords
skeleton
clothes
point
branches
symmetry
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201510237236.1A
Other languages
Chinese (zh)
Other versions
CN104952065B (en
Inventor
王贝
李基拓
曾继平
陈�光
陆国栋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201510237236.1A priority Critical patent/CN104952065B/en
Publication of CN104952065A publication Critical patent/CN104952065A/en
Application granted granted Critical
Publication of CN104952065B publication Critical patent/CN104952065B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Image Analysis (AREA)

Abstract

The invention discloses a method for building a multilayer detailed skeleton model of garment images. The method includes: for garment profile extracted from an image in advance, applying a constraint Delaunay trianglization method to extract a garment skeleton corresponding to the garment profile, and smoothing the extracted garment skeleton; extracting garment profile symmetric axis at the same time, utilizing the garment profile symmetric axis to match garment skeleton branches, and utilizing an improved webpage sequencing method to calculate importance values of skeleton key points; utilizing iterative loop to calculate importance values of the garment skeleton key points and simplify the garment skeleton to acquire the multilayer detailed skeleton model of final garment images. By the method, the garment skeleton can be simplified sequentially, and the multilayer detailed skeleton model of the garment images from simple to complex can be acquired.

Description

A kind of method setting up the multi-level details skeleton pattern of image of clothing
Technical field
The present invention relates to a kind of method setting up image of clothing skeleton pattern, especially a kind of method setting up the multi-level details skeleton pattern of image of clothing.
Background technology
In image processing field, skeleton can be described object by a kind of simply compact mode, and especially in the process application of image of clothing, skeleton can embody a concentrated reflection of a series of information of clothes, such as symmetry, directivity, attribute.But in actual applications, skeleton is not only easily subject to the interference of body outline and produces irrational tiny branch, and self also may comprise too much excessively assorted subbranch.Therefore, skeleton is simplified to have in the skeleton enriching details the important means extracting validity feature.Traditional skeleton short-cut method great majority take simple method of cutting out, an i.e. first selected parameter (usually choosing length is parameter) relevant to skeleton attribute, the threshold value then by arranging this parameter crops disturbs branch and unessential skeleton branches.But this method can only obtain the skeleton of a relative simplicity, and can not pull out most important skeleton prototype; In addition, the threshold value of parameter is not easy to determine, usually needs to carry out artificially arranging and adjusting.There is the shortcoming of uncertain large and inefficiency in this method, thus cannot meet skeleton in actual applications multiple, apply fast.
Summary of the invention
The shortcoming of and inefficiency large for the uncertainty existing for the above-mentioned traditional skeleton short-cut method mentioned, the invention provides a kind of method setting up the multi-level details skeleton pattern of image of clothing, by by the PageRank approach application of amendment in the skeleton setting up image of clothing, importance rate assessment is carried out to skeleton branches, and carry out the classification of different levels, automatically set up the model of a multi-level details.This model has great application prospect in clothes classification and clothing matching.
As shown in Figure 1, the technical solution adopted in the present invention comprises the following steps:
Step one, extract clothes skeletons: extract clothes profile and equidistant sampling obtains the polygon of clothes profile by OPENCV instrument from image of clothing, using the polygon of composition clothes profile as constraint condition, Delaunay triangulation is carried out to polygonal internal, make the triangle of polygonal internal keep the leg-of-mutton attribute of Delaunay, obtain the triangle with two kinds of dissimilar limits; Two kinds of dissimilar limits are the internal edges of the inside being present in clothes profile and the boundary edge be present on clothes profile, boundary edge is made up of the line between two adjacent clothes point, the triangle obtained by subdivision is again divided into three classes according to the quantity of had internal edges and boundary edge, extract each leg-of-mutton inner skeleton line segment, all leg-of-mutton inner skeleton line segment end points are joined end to end, obtains clothes skeleton;
In above-mentioned image of clothing image background color single and with clothing color there is obvious aberration, clothes profile can be demonstrated.
Step 2, carry out smooth to extracted clothes skeleton:
Extract the clothes skeleton branches in clothes skeleton, in order to smooth clothes skeleton, respectively Bezier curve is carried out to each clothes skeleton branches, using the whole story point of two of clothes skeleton branches end points as Bezier curve, the skeleton tie point in the middle of clothes skeleton branches is as the reference mark of Bezier curve; So both can reach smooth effect, the invariant position of key point can be kept again, ensure that the integral position of clothes skeleton was constant with this.
Step 3, extraction clothes profile axis of symmetry:
Major part clothes all have symmetry, therefore first use the equidistant sampled point of principal component analysis (PCA) PCA (Principal Component Analysis) method to clothes profile to carry out dimensionality reduction and obtain two proper vectors, two proper vectors are respectively as principal direction and time direction; With the focus point of clothes profile for through point, respectively with principal direction and time direction for rectilinear direction, composition two PCA axles separately, clothes contours segmentation is become outline line P and Q of both sides by each PCA axle;
With wherein any one PCA axle for mirror shaft, by the outline line P Mirroring Mapping of wherein side to opposite side, obtain mirrored portion outline line P ', this mirrored portion outline line P ' and the homonymy of outline line Q at mirror shaft originally splitting another part side obtained, and (mirror image Hausdorff distance table is shown as MHD (Mirror Hausdorff Distance) value to calculate the mirror image Hausdorff distance of mirror shaft homonymy two outline line P ' and Q, be used for weighing the symmetry being split two parts outline line P and Q obtained by PCA axle), calculate the mirror image Hausdorff distance that two PCA axle segmentations obtain both sides outline line respectively, choose less mirror image Hausdorff apart from corresponding PCA axle (namely having more the symmetric axle of profile) as initial clothes profile axis of symmetry l 0, initial clothes profile axis of symmetry l 0carry out iteration adjustment and obtain real garment profile axis of symmetry l,
Step 4, coupling clothes skeleton branches:
Utilize real garment profile axis of symmetry l to carry out left and right to clothes skeleton branches according to the position of center of gravity to sort out, calculate left and right side clothes skeleton branches mirror image Hausdorff distance between any two; Choose about two the clothes skeleton branches of mirror image Hausdorff distance corresponding to minimum value successively right as coupling, about two the clothes skeleton branches chosen as coupling is right do not mate right object as choosing next time, take until wherein the clothes skeleton branches of side is selected, obtain many group couplings right;
Mirror image Hausdorff distance is arranged successively from small to large, use large Tianjin Otsu method again, each mirror image Hausdorff distance is successively as partition value, with partition value, all mirror image Hausdorff distances obtained are divided into two classes according to size, and calculate the respective variance within clusters of two classes and inter-class variance, the partition value getting the minimum and maximum correspondence of inter-class variance of the variance within clusters of two classes, as optimal threshold, removes the mirror image Hausdorff being greater than optimal threshold right apart from corresponding clothes skeleton branches coupling; The mirror image Hausdorff being greater than optimal threshold mates normally some symmetry coupling that is not high, matching error is right apart from corresponding clothes skeleton branches, therefore finally removes these clothes skeleton branches coupling right.
The Web page sequencing method (PageRank method) that step 5, utilization are improved calculates the importance values of skeleton key point:
Identical importance initial value is given as the page node of Web page sequencing method using each skeleton key point, each clothes skeleton branches illustrates a link between two skeleton key points (two end points of clothes skeleton branches), comprises chain and enters chain;
All page nodes and their linking relationships each other constitute a network graphics drawing, utilize PageRank method can calculate the importance values of each page node in this network graphics drawing.PageRank method has just started to give each webpage identical importance values, by going out chain, entering the PageRank score (being defined as PR value) that the continuous iterative computation of the relation of chain upgrades each page node, until score is stablized.In the present invention, the PageRank method of improvement by going out chain, (i.e. the length of clothes skeleton branches) adds in method to enter the path distance of chain.
Using the distribution factor of the length of clothes skeleton branches as importance values in Web page sequencing method, the continuous iterative computation of the relation adopting Web page sequencing method to pass through chain, enter chain upgrades, until numerical stability obtains the importance values of last each page node, the following formula of concrete employing:
PR ( E ) = Σ i = 0 m - 1 L ( E , E i ) Σ j = 0 n - 1 L ( E i , E j ) PR ( E i )
Wherein, the importance values of PR (E) representation page node E, the page node E that m representation page node E links iquantity, N (E i) representation page node E ithe page number of nodes linked, n representation page node E ithe page node E linked jquantity, L (E i, E j) represent two page node E that interlink iand E jbetween path, the page node E that i representation page node E links iordinal number, j representation page node E ithe page node E linked jordinal number;
Step 6, utilize iterative loop to calculate the importance values of clothes skeletons, set up the multi-level details skeleton pattern of image of clothing:
Unique clothes skeleton branches that the skeleton tip point of clothes connects is defined as skeleton tip branch, the importance values of double counting clothes skeleton key point and simplify clothes skeleton, obtains the multi-level details skeleton pattern of last image of clothing.
From initial clothes skeleton, above process can obtain a series of clothes skeleton simplified successively, jointly constitutes a series of clothes skeletons from numerous to letter.In order to the convenience applied, the arrangement mode of these clothes skeletons again inverted order is arranged, obtains numerous a series of clothes skeletons of conforming to the principle of simplicity, i.e. the multi-level details skeleton pattern of image of clothing.
In described step one, the triangle that subdivision obtains is divided into three classes in the following ways according to the quantity of had internal edges and boundary edge: what have an internal edges and two boundary edge is I class triangle, what have two internal edges and a boundary edge is II class triangle, and what have three internal edges is III class triangle.
In described step one, the mid point of three internal edges and the respective line of triangle Voronoi point in line, III class triangle between the mid point that inner skeleton line segment comprises two internal edges in the line of the mid point of internal edges in I class triangle and the right triangular apex of this internal edges, II class triangle.
In described step 2, the clothes skeleton branches extracted in clothes skeleton is specially: the point forming clothes skeleton according to step one has following three classes: connect the skeleton point of crossing of three triangle interior skeleton line segments, the skeleton tie point connecting two triangle interior skeleton line segments and the skeleton tip point being only connected a triangle interior skeleton line segment; Using skeleton point of crossing and skeleton tip point as skeleton key point, if there is not skeleton point of crossing in the skeleton line segment path between two skeleton key points, then this skeleton line segment path is clothes skeleton branches, and these two skeleton key points are two end points of clothes skeleton branches.
In described step 3, to initial clothes profile axis of symmetry l kcarry out iteration adjustment in the following ways with close to real garment profile axis of symmetry:
3.1) clothes profile axis of symmetry l kbe the outline line P of both sides by clothes contours segmentation kand Q k, k is clothes profile axis of symmetry adjustment number of times, incites somebody to action wherein a part of outline line P kwith clothes profile axis of symmetry l kfor benchmark is mapped to opposite side, obtain mirrored portion profile P ' k;
3.2) by mirrored portion profile P ' kwith opposite side outline line Q kon o'clock as the set of two points, constantly calculate by ICP (Iterative Closest Points) method and upgrade and obtain mirrored portion profile P " k, the mirrored portion profile P after renewal " kwith its initial unmapped partial contour P kcarry out one_to_one corresponding a little, obtain each pair of corresponding match point;
3.3) utilize least square method, the mid point of often pair of match point connecting line segment is carried out matching as point set and obtains new clothes profile axis of symmetry l k+1;
3.4) repeat above 3.1), 3.2), 3.3) step carries out iteration adjustment, until the clothes profile axis of symmetry l that current procedures obtains nwith clothes profile axis of symmetry l obtained in the previous step n-1misalignment angle be less than 2 °, stop calculate, by the clothes profile axis of symmetry l calculated for the last time nas real garment profile axis of symmetry l, n by obtain real garment profile axis of symmetry the adjustment number of times of process.
In described step 4, left and right side clothes skeleton branches mirror image Hausdorff distance between any two comprises the mirror image Hausdorff distance of the arbitrary clothes skeleton branches in wherein side and each clothes skeleton branches of opposite side.
Described step 6 specifically comprises: the importance values 6.1) being calculated each skeleton key point by step 5 is mated each clothes skeleton branches obtained and mated right with step 4, calculate the mean value of skeleton tip branch coupling centering two skeleton tip point importance values;
6.2) the skeleton tip Point matching that the mean value selecting importance values is less than importance initial value to and the skeleton tip branch coupling removed corresponding to it is right; For the single skeleton tip branch of not mating, remove the skeleton tip branch that importance values is less than importance initial value, generate new clothes skeleton;
6.3) successively iteration repeat 6.1), 6.2) step, until last remaining clothes skeleton branches number is 1 or 0, and the importance values calculating new simplification clothes skeleton cannot be continued, then complete the multi-level details skeleton pattern setting up image of clothing.
The beneficial effect that the present invention has is:
The present invention by by the PageRank approach application of amendment in the skeleton setting up image of clothing, importance rate assessment is carried out to skeleton branches, and carries out the classification of different levels, automatically set up the model of a multi-level details.The present invention simplifies clothes skeleton successively according to significance level, has wide practical use in clothes identification and clothes match party mask.The inventive method efficiency is high, and determinacy is good, can meet skeleton in actual applications multiple, apply fast.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the method for the invention.
Fig. 2 is the clothes profile that the embodiment of the present invention is extracted.
Fig. 3 is the clothes skeleton that the embodiment of the present invention is extracted.
Fig. 4 is the clothes skeleton branches of the embodiment of the present invention.
Fig. 5 is the smooth clothes skeleton of the embodiment of the present invention.
Fig. 6 is computation process and the result of embodiment of the present invention clothes profile axis of symmetry.
Fig. 7 is that the coupling of embodiment of the present invention clothes skeleton branches is right.
Fig. 8 is the process that the embodiment of the present invention simplifies clothes skeleton and skeleton key point.
In figure: 1, clothes, 2, clothes profile, 3, internal edges, 4, boundary edge, 5, clothes skeleton, 6, skeleton point of crossing, 7, skeleton tie point, 8, skeleton tip point, 9, skeleton key point, 10, clothes skeleton branches, 11, clothes profile axis of symmetry, 12, PCA axle, 13, mate right, 14, multi-level details skeleton pattern, 15, skeleton tip branch.
Embodiment
Below in conjunction with drawings and Examples, the invention will be further described.
Specific embodiments of the invention are as follows:
Step one: as shown in Figure 2, extracts the clothes profile 2 of clothes 1 correspondence, and the polygon of composition clothes profile 2 is carried out Delaunay triangulation as constraint condition.
As shown in Fig. 3 (a), the triangle that subdivision obtains has internal edges 3 and boundary edge 4.The triangle obtained by subdivision is divided into following three classes according to the quantity of had internal edges 3 and boundary edge 4: what have an internal edges 3 and two boundary edge 4 is I class triangle; What have two internal edges 3 and a boundary edge 4 is II class triangle; What have three internal edges 3 is III class triangle.
Extract each leg-of-mutton inner skeleton line segment, i.e. the mid point of three internal edges and the respective line of triangle Voronoi point in the mid point line of two internal edges, III class triangle in the line of the mid point of internal edges and the right triangular apex of this internal edges, II class triangle in I class triangle.All leg-of-mutton inner skeleton line segment end points are joined end to end, clothes skeleton 5 can be obtained, as shown in Fig. 3 (b) He Fig. 3 (c).
Step 2: carry out smooth to extracted clothes skeleton 5.
As shown in Fig. 4 (b), the point of composition clothes skeleton 5 has three classes: connect the skeleton point of crossing 6 of three triangle interior skeleton line segments, the skeleton tie point 7 connecting two triangle interior skeleton line segments and the skeleton tip point 8 being only connected a triangle interior skeleton line segment.
By wherein skeleton point of crossing 6 and skeleton tip point 8 are defined as skeleton key point 9, if there is not skeleton point of crossing 6 in the skeleton line segment path between two skeleton key points 9, be then clothes skeleton branches 10 by this skeleton line segment path definition, these two skeleton key points 9 are exactly two end points of clothes skeleton branches 10, as shown in Fig. 4 (c).Using the whole story point of two of clothes skeleton branches 10 end points as Bezier curve, middle skeleton tie point is as the reference mark of Bezier curve, and carry out Bezier curve to each clothes skeleton branches 10, result is as shown in Fig. 5 (b).
Step 3: extract clothes profile axis of symmetry 11.
Carry out dimensionality reduction with the equidistant sampled point of principal component analysis (PCA) PCA (Principal Component Analysis) method to clothes profile and obtain two proper vectors, two proper vectors are respectively as principal direction and time direction.As shown in Figure 6, with the focus point of clothes profile 2 for through point, respectively with principal direction and time direction for rectilinear direction, composition two PCA axles 12 separately, clothes profile 2 is divided into two parts outline line P and Q by each PCA axle 12.With one of them PCA axle 12 for mirror shaft, incite somebody to action wherein a part of outline line E jmirroring Mapping is to opposite side, obtain mirrored portion outline line P ', this mirrored portion outline line P ' and originally split another part outline line Q of obtaining homonymy at mirror shaft, and calculate the mirror image Hausdorff distance of mirror shaft homonymy two outline line P ' and Q, be namely called MHD value.Calculate the MHD value that two PCA axles 12 split two parts outline line obtained respectively, choose and have PCA axle 12 corresponding to smaller value as initial clothes profile axis of symmetry l k(k is clothes profile axis of symmetry adjustment number of times).
To initial clothes profile axis of symmetry l kcarry out further iteration adjustment with close to real clothes profile axis of symmetry 11:
3.1) initial clothes profile axis of symmetry l kbe two parts outline line P by clothes contours segmentation kand Q k, incite somebody to action wherein a part of outline line l kwith initial clothes profile axis of symmetry l kfor benchmark is mapped to opposite side, obtain mirrored portion profile P ' k, as shown in Fig. 6 (a);
3.2) by mirrored portion profile P ' kwith another part outline line Q that segmentation obtains kon o'clock as the set of two points, constantly calculate by ICP method and upgrade this mirrored portion profile P " k, the mirrored portion profile P after renewal " kwith its initial unmapped partial contour P kcarry out one_to_one corresponding a little, as shown in Fig. 6 (b);
3.3) utilize least square method, the mid point of often pair of corresponding point is carried out matching as point set and obtains new clothes profile axis of symmetry l k+1, as shown in Fig. 6 (c);
3.4) repeat above 3.1), 3.2), 3.3) step, until the clothes profile axis of symmetry l obtained n(n by obtain real clothes profile axis of symmetry the adjustment number of times of process) with the clothes profile axis of symmetry l of previous step n-1misalignment angle be less than 2 °, stop calculate, by l nas real clothes profile axis of symmetry 11.
Step 4: utilize clothes profile axis of symmetry 11 pairs of clothes skeleton branches 10 to carry out left and right and sort out, as shown in Fig. 7 (a), calculate the value of MHD between two of left and right clothes skeleton branches 10, the value of MHD between two in the present embodiment is as shown in table 1, about two the clothes skeleton branches chosen successively corresponding to MHD value minimum value are right as coupling, about two the clothes skeleton branches chosen as coupling is right do not mate right object as choosing next time, take until wherein the clothes skeleton branches of side is selected, obtain many group couplings right:
Table 1
The result obtained according to table 1 is as shown in Fig. 7 (b) to Fig. 7 (m).Use large Tianjin Otsu method, MHD value is arranged successively from small to large, choose each MHD value respectively as threshold value, coupling is divided into two classes to 13 according to the MHD value size of correspondence by this threshold value, when the variance within clusters of this two class is minimum and inter-class variance is maximum, obtain the optimal threshold 20.082 of the present embodiment.Be greater than the coupling 13 of the clothes skeleton branches corresponding to MHD value of optimal threshold to normally some symmetry coupling that is not high, matching error is right, according to the result in table 1, be (l) and (m) in Fig. 7, finally remove these couplings to 13.
Step 5: utilize the PageRank method improved to calculate the importance values of skeleton key point 9, be PR value.
Using the page node of skeleton key point 9 as PageRank method, each clothes skeleton branches 10 illustrates a link between two skeleton key points 9 (two end points of clothes skeleton branches 10).All page nodes and their linking relationships each other constitute a network graphics drawing.The PageRank method formula improved is:
PR ( E ) = Σ i = 0 m - 1 L ( E , E i ) Σ j = 0 n - 1 L ( E i , E j ) PR ( E i )
The PR value of PR (E) representation page node E, the page node E that m representation page node E links iquantity, N (E i) representation page node E ithe page number of nodes linked, n representation page node E ithe page node E linked jquantity, L (E i, E j) represent two page node E that interlink iand E jbetween path.
In the present embodiment, give identical PR initial value 1.0 to each skeleton key point 9, by the PageRank formula iterative computation after improvement until numerical stability, obtain the network PR value of this clothes skeleton.
Step 6: utilize iterative loop to calculate the network PR value of clothes skeleton, set up the multi-level details skeleton pattern 14 of image of clothing.
Unique clothes skeleton branches 10 that the skeleton tip point 8 of clothes connects is defined as skeleton tip branch 15, the PR value of double counting clothes skeleton key point 9 and simplify clothes skeleton 5, obtains the multi-level details skeleton pattern 14 of last image of clothing:
6.1) calculate the PR value of the skeleton key point 9 of clothes according to step 5, and mate each clothes skeleton branches 10 according to step 4, calculate the PR mean value that the corresponding skeleton tip point 8 of centering is mated in skeleton tip branch 15;
6.2) select skeleton tip branch 15 to mate skeleton tip point 8 that right mean P R value is less than initial value 1.0 and mate right, the skeleton tip branch 15 cropped corresponding to it mates right; For the single skeleton tip branch 15 of not mating, the PR value directly removing skeleton tip point 8 is less than the skeleton tip branch 15 of 1.0, generates a new simplification clothes skeleton.
6.3) 6.1 are repeated to the simplification clothes skeleton that newly obtains), 6.2) step, until last clothes skeleton branches 10 number residue 1 or 0, the PR value that the present embodiment respectively walks is as shown in table 2, network graphics drawing as shown in Figure 8,1 result of calculation corresponding diagram 8 (a) in table 2,2 result of calculation corresponding diagram 8 (b), 3 result of calculation corresponding diagram 8 (c).
Table 2
From initial clothes skeleton 5, above process can obtain a series of clothes skeleton simplified successively, namely obtains the multi-level details skeleton pattern 14 of final image of clothing.
As can be seen here, the present invention realizes the model establishing a multi-level details, and simplify successively clothes skeleton according to significance level, efficiency is high, and determinacy is good, can be used for the aspects such as clothes identification and clothing matching.
Above-mentioned embodiment is used for explaining and the present invention is described, instead of limits the invention, and in the protection domain of spirit of the present invention and claim, any amendment make the present invention and change, all fall into protection scope of the present invention.

Claims (7)

1. set up a method for the multi-level details skeleton pattern of image of clothing, it is characterized in that:
Step one, extraction clothes skeleton:
Clothes profiles is extracted and equidistant sampling obtains the polygon of clothes profile from image of clothing by OPENCV instrument, using the polygon of composition clothes profile as constraint condition, Delaunay triangulation is carried out to polygonal internal, make the triangle of polygonal internal keep the leg-of-mutton attribute of Delaunay, obtain the triangle with two kinds of dissimilar limits; Two kinds of dissimilar limits are the internal edges of the inside being present in clothes profile and the boundary edge be present on clothes profile, the triangle obtained by subdivision is again divided into three classes according to the quantity of had internal edges and boundary edge, extract each leg-of-mutton inner skeleton line segment, all leg-of-mutton inner skeleton line segment end points are joined end to end, obtains clothes skeleton;
Step 2, carry out smooth to extracted clothes skeleton:
Extract the clothes skeleton branches in clothes skeleton, respectively Bezier curve is carried out to each clothes skeleton branches, using the whole story point of two of clothes skeleton branches end points as Bezier curve, the skeleton tie point in the middle of clothes skeleton branches is as the reference mark of Bezier curve;
Step 3, extraction clothes profile axis of symmetry:
Carry out dimensionality reduction with the equidistant sampled point of principal component analysis (PCA) PCA method to clothes profile and obtain two proper vectors, two proper vectors are respectively as principal direction and time direction; With the focus point of clothes profile for through point, respectively with principal direction and time direction for rectilinear direction, composition two PCA axles separately, clothes contours segmentation is become outline line P and Q of both sides by each PCA axle;
With wherein any one PCA axle for mirror shaft, by the outline line P Mirroring Mapping of wherein side to opposite side, obtain mirrored portion outline line P ', and calculate the mirror image Hausdorff distance of mirror shaft homonymy two outline line P ' and Q, calculate the mirror image Hausdorff distance that two PCA axle segmentations obtain both sides outline line respectively, choose less mirror image Hausdorff apart from corresponding PCA axle as initial clothes profile axis of symmetry l 0, initial clothes profile axis of symmetry l 0carry out iteration adjustment and obtain real garment profile axis of symmetry l;
Step 4, coupling clothes skeleton branches:
Utilize real garment profile axis of symmetry l to carry out left and right to clothes skeleton branches according to the position of center of gravity to sort out, calculate left and right side clothes skeleton branches mirror image Hausdorff distance between any two; Choose about two the clothes skeleton branches of mirror image Hausdorff distance corresponding to minimum value successively right as coupling, about two the clothes skeleton branches chosen as coupling is right do not mate right object as choosing next time, take until wherein the clothes skeleton branches of side is selected, obtain many group couplings right;
Mirror image Hausdorff distance is arranged successively from small to large, use large Tianjin Otsu method again, each mirror image Hausdorff distance is successively as partition value, with partition value, all mirror image Hausdorff distances obtained are divided into two classes according to size, and calculate the respective variance within clusters of two classes and inter-class variance, the partition value getting the minimum and maximum correspondence of inter-class variance of the variance within clusters of two classes, as optimal threshold, removes the mirror image Hausdorff being greater than optimal threshold right apart from corresponding clothes skeleton branches coupling;
The Web page sequencing method that step 5, utilization are improved calculates the importance values of skeleton key point:
Give identical importance initial value using each skeleton key point as the page node of Web page sequencing method, each clothes skeleton branches illustrates a link between two skeleton key points, comprises chain and enters chain; Using the distribution factor of the length of clothes skeleton branches as importance values in Web page sequencing method, the continuous iterative computation of the relation adopting Web page sequencing method to pass through chain, enter chain upgrades, until numerical stability obtains the importance values of last each page node, the following formula of concrete employing:
PR ( E ) = Σ i = 0 m - 1 L ( E , E i ) Σ j = 0 n - 1 L ( E i , E j ) PR ( E i )
Wherein, the importance values of PR (E) representation page node E, the page node E that m representation page node E links iquantity, N (E i) representation page node E ithe page number of nodes linked, n representation page node E ithe page node E linked jquantity, L (E i, E j) represent two page node E that interlink iand E jbetween path, the page node E that i representation page node E links iordinal number, j representation page node E ithe page node E linked iordinal number;
Step 6, utilize iterative loop to calculate the importance values of clothes skeletons, set up the multi-level details skeleton pattern of image of clothing:
Unique clothes skeleton branches that the skeleton tip point of clothes connects is defined as skeleton tip branch, the importance values of double counting clothes skeleton key point and simplify clothes skeleton, obtains the multi-level details skeleton pattern of last image of clothing.
2. a kind of method setting up the multi-level details skeleton pattern of image of clothing according to claim 1, it is characterized in that: in described step one, the triangle that subdivision obtains is divided into three classes in the following ways according to the quantity of had internal edges and boundary edge: what have an internal edges and two boundary edge is I class triangle, what have two internal edges and a boundary edge is II class triangle, and what have three internal edges is III class triangle.
3. a kind of method setting up the multi-level details skeleton pattern of image of clothing according to claim 1, it is characterized in that: in described step one, the mid point of three internal edges and the respective line of triangle Voronoi point in line, III class triangle between the mid point that inner skeleton line segment comprises two internal edges in the line of the mid point of internal edges in I class triangle and the right triangular apex of this internal edges, II class triangle.
4. a kind of method setting up the multi-level details skeleton pattern of image of clothing according to claim 1, it is characterized in that: in described step 2, the clothes skeleton branches extracted in clothes skeleton is specially: the point forming clothes skeleton according to step one has following three classes: connect the skeleton point of crossing of three triangle interior skeleton line segments, the skeleton tie point connecting two triangle interior skeleton line segments and the skeleton tip point being only connected a triangle interior skeleton line segment; Using skeleton point of crossing and skeleton tip point as skeleton key point, if there is not skeleton point of crossing in the skeleton line segment path between two skeleton key points, then this skeleton line segment path is clothes skeleton branches, and these two skeleton key points are two end points of clothes skeleton branches.
5. a kind of method setting up the multi-level details skeleton pattern of image of clothing according to claim 1, is characterized in that: in described step 3, to initial clothes profile axis of symmetry l 0carry out iteration adjustment in the following ways with close to real garment profile axis of symmetry:
3.1) clothes profile axis of symmetry l kbe the outline line P of both sides by clothes contours segmentation kand Q k, k is clothes profile axis of symmetry adjustment number of times, incites somebody to action wherein a part of outline line P kwith clothes profile axis of symmetry l kfor benchmark is mapped to opposite side, obtain mirrored portion profile P ' k;
3.2) by mirrored portion profile P ' kwith opposite side outline line Q kon o'clock as the set of two points, constantly calculate by ICP (Iterative Closest Points) method and upgrade and obtain mirrored portion profile P " k, the mirrored portion profile P after renewal " kwith its initial unmapped partial contour P kcarry out one_to_one corresponding a little, obtain each pair of corresponding match point;
3.3) utilize least square method, the mid point of often pair of match point connecting line segment is carried out matching as point set and obtains new clothes profile axis of symmetry l k+1;
3.4) repeat above 3.1), 3.2), 3.3) step carries out iteration adjustment, until the clothes profile axis of symmetry l that current procedures obtains nwith clothes profile axis of symmetry l obtained in the previous step n-1misalignment angle be less than 2 °, stop calculate, by the clothes profile axis of symmetry l calculated for the last time nas real garment profile axis of symmetry l, n by obtain real garment profile axis of symmetry the adjustment number of times of process.
6. a kind of method setting up the multi-level details skeleton pattern of image of clothing according to claim 1, it is characterized in that: in described step 4, left and right side clothes skeleton branches mirror image Hausdorff distance between any two comprises the mirror image Hausdorff distance of the arbitrary clothes skeleton branches in wherein side and each clothes skeleton branches of opposite side.
7. a kind of method setting up the multi-level details skeleton pattern of image of clothing according to claim 1, is characterized in that: described step 6 specifically comprises:
6.1) importance values being calculated each skeleton key point by step 5 is mated each clothes skeleton branches obtained and is mated right with step 4, calculate the mean value of skeleton tip branch coupling centering two skeleton tip point importance values;
6.2) the skeleton tip Point matching that the mean value selecting importance values is less than importance initial value to and the skeleton tip branch coupling removed corresponding to it is right; For the single skeleton tip branch of not mating, remove the skeleton tip branch that importance values is less than importance initial value, generate new clothes skeleton;
6.3) successively iteration repeat 6.1), 6.2) step, until last remaining clothes skeleton branches number is 1 or 0, and the importance values calculating new simplification clothes skeleton cannot be continued, then complete the multi-level details skeleton pattern setting up image of clothing.
CN201510237236.1A 2015-05-10 2015-05-10 A kind of method of setting up the multi-level details skeleton pattern of image of clothing Active CN104952065B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510237236.1A CN104952065B (en) 2015-05-10 2015-05-10 A kind of method of setting up the multi-level details skeleton pattern of image of clothing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510237236.1A CN104952065B (en) 2015-05-10 2015-05-10 A kind of method of setting up the multi-level details skeleton pattern of image of clothing

Publications (2)

Publication Number Publication Date
CN104952065A true CN104952065A (en) 2015-09-30
CN104952065B CN104952065B (en) 2016-05-18

Family

ID=54166693

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510237236.1A Active CN104952065B (en) 2015-05-10 2015-05-10 A kind of method of setting up the multi-level details skeleton pattern of image of clothing

Country Status (1)

Country Link
CN (1) CN104952065B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106384126A (en) * 2016-09-07 2017-02-08 东华大学 Clothes pattern identification method based on contour curvature feature points and support vector machine
WO2017107324A1 (en) * 2015-12-24 2017-06-29 中兴通讯股份有限公司 Photographing mode processing method and device
CN110378959A (en) * 2019-07-15 2019-10-25 杭州恢弘科技有限公司 A kind of clothes auxiliary print is boiling hot to position setting method, localization method and auxiliary print ironing process
CN112419444A (en) * 2020-12-09 2021-02-26 北京维盛视通科技有限公司 Clothing sheet drawing method and device, electronic equipment and storage medium
CN113554758A (en) * 2021-07-21 2021-10-26 南京师范大学 Voronoi diagram-based parallel contour line branch processing method and device

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110169838A1 (en) * 2007-06-01 2011-07-14 Branets Larisa V Generation of Constrained Voronoi Grid In A Plane
CN102254343A (en) * 2011-07-01 2011-11-23 浙江理工大学 Convex hull and OBB (Oriented Bounding Box)-based three-dimensional grid model framework extracting method
CN103400372A (en) * 2013-07-10 2013-11-20 中国科学技术大学 Three-dimensional topological information extraction method based on Reeb graph description

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110169838A1 (en) * 2007-06-01 2011-07-14 Branets Larisa V Generation of Constrained Voronoi Grid In A Plane
CN102254343A (en) * 2011-07-01 2011-11-23 浙江理工大学 Convex hull and OBB (Oriented Bounding Box)-based three-dimensional grid model framework extracting method
CN103400372A (en) * 2013-07-10 2013-11-20 中国科学技术大学 Three-dimensional topological information extraction method based on Reeb graph description

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
宫法明: "一种改进的基于特征点求解的骨架提取算法", 《微型电脑应用》, vol. 26, no. 4, 30 April 2010 (2010-04-30) *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017107324A1 (en) * 2015-12-24 2017-06-29 中兴通讯股份有限公司 Photographing mode processing method and device
CN106384126A (en) * 2016-09-07 2017-02-08 东华大学 Clothes pattern identification method based on contour curvature feature points and support vector machine
CN110378959A (en) * 2019-07-15 2019-10-25 杭州恢弘科技有限公司 A kind of clothes auxiliary print is boiling hot to position setting method, localization method and auxiliary print ironing process
CN112419444A (en) * 2020-12-09 2021-02-26 北京维盛视通科技有限公司 Clothing sheet drawing method and device, electronic equipment and storage medium
CN112419444B (en) * 2020-12-09 2024-03-29 北京维盛视通科技有限公司 Clothing sheet drawing method and device, electronic equipment and storage medium
CN113554758A (en) * 2021-07-21 2021-10-26 南京师范大学 Voronoi diagram-based parallel contour line branch processing method and device
CN113554758B (en) * 2021-07-21 2024-08-23 南京师范大学 Parallel contour line branch processing method and device based on Voronoi diagram

Also Published As

Publication number Publication date
CN104952065B (en) 2016-05-18

Similar Documents

Publication Publication Date Title
CN104952065A (en) Method for building multilayer detailed skeleton model of garment images
CN102004922B (en) High-resolution remote sensing image plane extraction method based on skeleton characteristic
Gyulassy et al. Efficient computation of Morse-Smale complexes for three-dimensional scalar functions
Chung et al. ω-harmonic functions and inverse conductivity problems on networks
CN103247041B (en) A kind of dividing method of the cloud data of the many geometric properties based on local sampling
WO2019203231A1 (en) Three-dimensional point cloud label learning device, three-dimensional point cloud label estimating device, three-dimensional point cloud label learning method, three-dimensional point cloud label estimating method, and program
US9741128B2 (en) Method and system for characterizing plan phenotype
CN106533742B (en) Weighting directed complex networks networking method based on time sequence model characterization
WO2018219522A1 (en) Method and apparatus for producing a lane-accurate road map
CN103268631A (en) Method and device for extracting point cloud framework
CN103400372A (en) Three-dimensional topological information extraction method based on Reeb graph description
CN104346481A (en) Community detection method based on dynamic synchronous model
De Runz et al. Unsupervised visual data mining using self-organizing maps and a data-driven color mapping
CN104657986A (en) Quasi-dense matching extension method based on subspace fusion and consistency constraint
KR20150089663A (en) Device for multi-shape primitives fitting of 3D point clouds using graph-based segmentation and method thereof
CN105045863A (en) Method and system used for entity matching
CN103955950A (en) Image tracking method utilizing key point feature matching
Alahakoon et al. A self-growing cluster development approach to data mining
CN113514072B (en) Road matching method oriented to navigation data and large-scale drawing data
Bertault et al. Drawing hypergraphs in the subset standard (short demo paper)
Zeng et al. A distance-based parameter free algorithm for curve reconstruction
CN108470094B (en) Truss structure three-dimensional model intelligent generation method
CN103985149B (en) Method for describing point features of three-dimensional colorful point cloud
CN115544307A (en) Directed graph data feature extraction and expression method and system based on incidence matrix
CN109765588B (en) Sparse track smooth error correction system and method

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant