CN106780506A - A kind of interactive image segmentation method based on multi-source shortest path distance - Google Patents
A kind of interactive image segmentation method based on multi-source shortest path distance Download PDFInfo
- Publication number
- CN106780506A CN106780506A CN201611039325.6A CN201611039325A CN106780506A CN 106780506 A CN106780506 A CN 106780506A CN 201611039325 A CN201611039325 A CN 201611039325A CN 106780506 A CN106780506 A CN 106780506A
- Authority
- CN
- China
- Prior art keywords
- pixel
- mark
- image
- unlabelled
- split
- 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
Links
- 238000003709 image segmentation Methods 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 26
- 230000002452 interceptive effect Effects 0.000 title claims abstract description 22
- 239000003550 marker Substances 0.000 claims abstract description 8
- 238000004364 calculation method Methods 0.000 claims abstract description 5
- 238000000638 solvent extraction Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 50
- 230000011218 segmentation Effects 0.000 claims description 16
- 238000004422 calculation algorithm Methods 0.000 description 28
- 230000006870 function Effects 0.000 description 18
- 238000005295 random walk Methods 0.000 description 11
- 230000001186 cumulative effect Effects 0.000 description 7
- 241000219109 Citrullus Species 0.000 description 5
- 235000012828 Citrullus lanatus var citroides Nutrition 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 241000234295 Musa Species 0.000 description 3
- 235000018290 Musa x paradisiaca Nutrition 0.000 description 3
- 238000000151 deposition Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 238000002203 pretreatment Methods 0.000 description 2
- 241000196324 Embryophyta Species 0.000 description 1
- 241000220225 Malus Species 0.000 description 1
- 235000011430 Malus pumila Nutrition 0.000 description 1
- 235000015103 Malus silvestris Nutrition 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 239000002304 perfume Substances 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Landscapes
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of interactive image segmentation method based on multi-source shortest path distance, including：Side right weight between S1, the pixel in the pixel value feature calculation image of image, and image is converted into by undirected weighted graph according to the side right between pixel again；S2, interior of articles each to be split in the picture mark multiple pixels as the mark point of each object to be split respectively, and the mark point of object to be split carries the respective mark of object to be split；S3, the mark point closest with it is found according to undirected weighted graph to all unlabelled pixels in image, multi-source Shortest Path Searching problem is converted into so as to divide the image into problem；S4, each unlabelled pixel is marked with the mark of the mark point closest with each unlabelled pixel, obtain the mark that each pixel in image is carried, and it is partitioning boundary that will carry the boundary marker of the pixel of different identification.The present invention can reach more preferably, high-quality image segmentation result.
Description
Technical field
The present invention relates to technical field of image segmentation.More particularly, to a kind of friendship based on multi-source shortest path distance
Mutual formula image partition method.
Background technology
Image segmentation is, according to information such as the texture between pixel in image, gray scale and colors, pixel to be sorted out.Figure
As segmentation is usually as the pre-treatment step of the technologies such as image procossing, pattern-recognition, computer vision, largely can be right
Accuracy rate or fineness of final objective result etc. produce important influence.In the last few years, due to the flourishing hair of above-mentioned technology
Exhibition, image segmentation is increasingly taken seriously.
However, due to the complexity and the subjectivity of segmentation of image, full automatic image Segmentation Technology is learned in the last few years
One of persons are considered what is be difficult to, and this is the reason for also exactly interactive image segmentation algorithm occupies dominant position always.
In Interactive Segmentation algorithm, the two classes classics image segmentation algorithm for occupying leading position is：Figure cut (graph cuts) algorithm and
Random walk (random walks) algorithm.
Algorithm is cut for figure, its main thought is that foreground pixel seed point and background pixel seed point are considered as into source/remittance section
Point, and then divide the image into problem and be converted into max-flow min-cut problem.However, due to this kind of " minimal cut " thought to a certain degree
On can find minimum boundary length, so as to " shortcut " problem can be triggered in practice.Also, figure cuts algorithm for noise
Robustness it is not strong, need to do original image some pretreatments under normal circumstances to reduce noise to final classification result
Interference.
For Random Walk Algorithm, its main thought is to seek the transition probability in image between adjacent pixel first, and is led to
Random Walk Algorithm is crossed, is that each pixel do not demarcated finds maximum probability up to terminal, so as to realize image segmentation purpose.
However, a total problem existing for such image segmentation algorithm based on random walk is：It can by balance background with
Gap between prospect and cause to the undesirable of image detail segmentation.
Also, the setting of existing its side right value of the image segmentation algorithm based on graph theory is class Gaussian function, although
Risk of " cumulative effect " segmentation wrong for remote pixel can be reduced to a certain extent, however, its but often cause by
In computational accuracy other erroneous segmentation result is produced beyond operating range.In actual applications, photoshop is carried for user
The stingy figure instrument based on smart scissors and based on threshold value is supplied.Wherein smart scissors scratch figure process to be needed necessarily
The manual intervention of amount, it is necessary to position image border to be split exactly.Stingy figure instrument range of application based on threshold value is narrow, causes
Result can not meet the application demand of most users.
Accordingly, it is desirable to provide a kind of interactive image segmentation that is quick, accurate, being easily based on multi-source shortest path distance
Method.
The content of the invention
It is an object of the invention to provide a kind of interactive image segmentation method based on multi-source shortest path distance, for
Efficiency present in existing image partition method is low, accuracy rate is low, there are problems that " cumulative effect.
To reach above-mentioned purpose, the present invention uses following technical proposals：
A kind of interactive image segmentation method based on multi-source shortest path distance, comprises the following steps：
Side right weight between S1, the pixel in the pixel value feature calculation image of image, and according to pixel it
Between side right image is converted into undirected weighted graph again；
S2, interior of articles each to be split in the picture mark multiple pixels as the mark of each object to be split respectively
Point, the mark point of object to be split carries the respective mark of object to be split；
S3, the mark point closest with it is found according to undirected weighted graph to all unlabelled pixels in image,
Multi-source Shortest Path Searching problem is converted into so as to divide the image into problem；
S4, the mark each unlabelled pixel of mark with the mark point closest with each unlabelled pixel, obtain
The mark that each pixel is carried in image, and be partitioning boundary by the boundary marker of the pixel with different identification.
Preferably, step S1 further includes following sub-step：
S1.1, the reachable domain that calculating side right weight is set；
The side between the pixel in each pixel and the reachable domain centered on each pixel in S1.2, calculating image
Weight, computing formula is as follows：
In formula, wi,jIt is the side right weight between pixel i and pixel j；IiThe gray scale of pixel i is represented, δ is first
Sensitive factor, n is the second sensitive factor；
S1.3, image is converted into by undirected weighted graph according to the side right between pixel again.
Preferably, up to domain be set to 5~20 close on it is reachable.
Preferably, IiIt is the rgb value of pixel, gray value or texture eigenvalue.
Preferably, the second sensitive factor n is set to 1, and the first sensitive factor δ is set to 1 if image is gray level image,
The first sensitive factor δ is set to 9 if image is coloured image.
Preferably, the mark point in step S2 be mark multiple discontinuous pixel or mark continuous lines in
Multiple continuous pixels.
Preferably, step S3 further includes following sub-step：
S3.1, the distance matrix that infinity is with the initial value of the size such as undirected weighted graph and each element is set up, and built
Stand for deposit final segmentation result and be with the initial value of the size such as distance matrix and each element 0 matrix of consequence；
S3.2, the correspondence position by the mark point of each object to be split in undirected weighted graph in distance matrix be entered as 0,
Correspondence position of the mark point of each object to be split in undirected weighted graph in matrix of consequence is entered as and each object to be split
The mark point corresponding numerical value of respective mark；
S3.3, the tax using the correspondence position apart from renewal function each unlabelled pixel of renewal in distance matrix
Value, when unlabelled pixel is in the assignment before the assignment of the correspondence position in distance matrix is less than its renewal after renewal, connects
The assignment of correspondence position of the unlabelled pixel in distance matrix after being updated, and by the unlabelled pixel in result
The assignment of the correspondence position in matrix is updated to the mark point represented in the side right weight in renewal function in matrix of consequence
Correspondence position assignment；
It is as follows apart from renewal function：
dist(u′new)=dist (u ') * θ+wu,u′
In formula, dist (u ') represents the assignment of correspondence positions of the unlabelled pixel u ' in distance matrix dist, θ
It is forgetting factor, wu,u′Represent the side right weight between unlabelled pixel u ' and mark point u, dist (u 'new) represent update after
Correspondence positions of the unlabelled pixel u ' in distance matrix dist assignment.
Preferably, forgetting factor
Beneficial effects of the present invention are as follows：
Technical scheme of the present invention compares the stingy figure instrument in the softwares such as nowadays more popular photoshop, no
Only faster, and interactive mode is more friendly, controllability strong, can reach in most cases more preferably for arithmetic speed
, high-quality image segmentation result.Specially：1. operational efficiency is superior to traditional figure and cuts and Random Walk Algorithm；2. transport
The row time is only relevant with image size, and how many unrelated with seed point (such as foreground point, background dot) is set in interactive process；
3. Segmentation of Multi-target purpose can be realized, and Riming time of algorithm is also only relevant with image size；4. can be prevented effectively from
Due to the over-segmentation problem that noise causes；5. compared to more traditional different, institute of the present invention with class Gaussian function statement side right value
Stating technical scheme can effectively eliminate " accumulation effect by introducing forgetting factor on the premise of not past Computing accuracy
Should ".
Brief description of the drawings
Specific embodiment of the invention is described in further detail below in conjunction with the accompanying drawings；
Fig. 1 shows a kind of interactive image segmentation method based on multi-source shortest path distance.
Specific embodiment
In order to illustrate more clearly of the present invention, the present invention is done further with reference to preferred embodiments and drawings
It is bright.Similar part is indicated with identical reference in accompanying drawing.It will be appreciated by those skilled in the art that institute is specific below
The content of description is illustrative and be not restrictive, and should not be limited the scope of the invention with this.
As shown in figure 1, a kind of interactive image segmentation method based on multi-source shortest path distance that the present embodiment is provided,
Comprise the following steps：
Side right weight between S1, the pixel in the pixel value feature calculation image of image, and according to pixel it
Between side right image is converted into undirected weighted graph again；
S2, interior of articles each to be split in the picture mark multiple pixels as the mark of each object to be split respectively
Point, the mark point of object to be split carries the respective mark of object to be split；
S3, the mark point closest with it is found according to undirected weighted graph to all unlabelled pixels in image,
Multi-source Shortest Path Searching problem is converted into so as to divide the image into problem；
S4, the mark each unlabelled pixel of mark with the mark point closest with each unlabelled pixel, obtain
The mark that each pixel is carried in image；It is partitioning boundary by the boundary marker of the pixel with different identification, completes
Image segmentation.
In the present embodiment, step S1 is the base of the entirely interactive image segmentation method based on multi-source shortest path distance
Plinth, step S1 further includes following sub-step：
S1.1, the reachable domain that calculating side right weight is set；
The side between the pixel in each pixel and the reachable domain centered on each pixel in S1.2, calculating image
Weight, computing formula is as follows：
In formula (1), wi,jIt is the side right weight between pixel i and pixel j；IiThe gray scale of pixel i is represented, for
Different image types, IiIt is the rgb value of pixel, gray value or texture eigenvalue；δ is the first sensitive factor, and its value is bigger
Represent the present embodiment and more insensitive is changed to pixel grey scale, n is the second sensitive factor, and it equally controls the present embodiment to pixel
The susceptibility of gray scale, but the influence of its susceptibility to pixel grey scale is index, the present embodiment set second it is sensitive because
The purpose of sub- n is to eliminate " cumulative effect "；
S1.3, image is converted into by undirected weighted graph according to the side right between pixel again.
In the present embodiment, set according to the size of image and actual segmentation demand up to domain, be usually arranged as 5~20 and close on
It is reachable, be specifically configured in the present embodiment 8 close on it is reachable, i.e., each pixel only with its around 8 pixels closing on calculate
Side right weight between pixel, and infinity is disposed as again with the side right of rest of pixels point.
In the present embodiment, the first sensitive factor δ is set to 1 if image is gray level image, if image is cromogram
As then the first sensitive factor δ is set to 9, and in order to not make distance function exceed Computing precision (exp (2552)=∞),
Second sensitive factor n is set to 1.It should be noted that " -1 " in formula (1) is particularly critical, due to exp, the former evaluation is equal
More than or equal to 1, in order to avoid " cumulative effect ", the present embodiment specifies have between same color or the adjacent pixel of gray scale
Side right weight is 0, represents and can freely reach between them.So-called " cumulative effect ", refers in the present embodiment in the present embodiment
The step of S3 in, when certain unlabelled pixel and certain mark point physical distance too far when, by have accumulated institute in path
Some distances, it is likely that this distance is more than apart from this near another mark point of certain unlabelled pixel physical distance, causes
Some remote unlabelled pixels are made to be labeled the mark of mistake in step s 4.Therefore, distance is put in the present embodiment
Greatly, i.e., determined according to formula (1) up to the weight between the pixel in domain.
Step S1 in the present embodiment solves the problems, such as to carry out undirected weighted graph modeling from gray scale or coloured image.This
Step is often the initial step of all image segmentation steps, and the present embodiment is used from unlike other image segmentation steps
Exponential type function represents the Gaussian function used in the distance between adjacent pixel, rather than other image segmentation steps.Get over phase
As adjacent pixel distance nearer rather than reachable probability it is bigger.Also, algorithm is cut by sensitive factor n with Random Walk Algorithm and figure
Be set to it is 2 different, the present embodiment in order to not cause that distance function exceedes the limitation of Computing precision, by the second sensitive factor n
It is set to 1.
In the present embodiment, the object to be split at least two in step S2, object to be split may include foreground object and
Background object.
In the present embodiment, " interior of articles each to be split in the picture marks multiple pixels as each to step S2 respectively
The mark point of object to be split, the mark point of object to be split carries the respective mark of object to be split " for example：Treating in image
Segmentation object include foreground object and background object, foreground object inner marker multiple pixels as foreground object mark
Point, the mark point mark A with foreground object of foreground object；In background object inner marker multiple pixel as background
The mark point of object, the mark point mark B with background object of background object.Or, for example：Thing to be split in image
Body includes apple, banana and watermelon, marks multiple pixels as the mark point of apple in apple internal, and the mark point of apple is equal
Mark C with apple；Banana inner marker multiple pixels as apple mark point, the mark point of banana is with perfume
The mark D of any of several broadleaf plants；Watermelon inner marker multiple pixels as watermelon mark point, the mark point of the watermelon mark with watermelon
Know E.
In the present embodiment, the mark point in step S2 can be the multiple discontinuous pixel of mark, or mark
Continuous lines in the continuous pixels of multiple.
In the present embodiment, the interior of articles each to be split in the picture in step S2 marks multiple pixel conducts respectively
The mark point of each object to be split can be performed by user using interactive device (such as mouse), what user can be clicked on mouse
Form is marked, it is also possible to pins left mouse button and is pulled to mark continuous lines.Additionally, user may be used also in the present embodiment
Segmentation operation is carried out with to multiple target simultaneously.
In the present embodiment, step S2 can be performed a plurality of times, you can repeatedly be marked with treating segmentation object, and can be preceding
It is marked again on the basis of segmentation result, is not limited in once.
In the present embodiment, step S3 further includes following sub-step：
S3.1, set up the distance matrix that infinitely great (inf) is with the initial value of the size such as undirected weighted graph and each element
Dist, and set up for depositing final segmentation result and being 0 with the initial value of the size such as distance matrix dist and each element
Matrix of consequence label；
S3.2, the correspondence position assignment by the mark point of each object to be split in undirected weighted graph in distance matrix dist
For the 0, correspondence position by the mark point of each object to be split in undirected weighted graph in matrix of consequence label is entered as and respectively treats
Split the corresponding numerical value of the respective mark of mark point of object；
S3.3, update the correspondence position of each unlabelled pixel in distance matrix dist using apart from renewal function
Assignment, the assignment of correspondence position of the unlabelled pixel in distance matrix dist is less than the assignment before its renewal after renewal
When, the assignment of correspondence position of the unlabelled pixel in distance matrix dist after receiving to update, and by the unlabelled picture
The assignment of correspondence position of the vegetarian refreshments in matrix of consequence label is updated to the side right weight w in renewal functionu,u′In mark
The assignment of correspondence positions of the point u in matrix of consequence label.
It is as follows apart from renewal function：
dist(u′new)=dist (u ') * θ+wu,u′ (2)
In formula (2), dist (u ') represents the tax of correspondence positions of the unlabelled pixel u ' in distance matrix dist
Value, θ is forgetting factor, wu,u′Represent the side right weight between unlabelled pixel u ' and mark point u, dist (u 'new) represent more
The assignment of the correspondence position of unlabelled pixel u ' after new in distance matrix dist.
Above-mentioned steps S3.1 is as follows to the false code of step S3.3, wherein, step S3.1 correspondence false codes are 1-2
OK, the row of false code the 3rd represents that it is empty list to set up for depositing the mark point and original state of each object to be split
Openlist, the row of false code the 4th is represented and utilizes SallMark point to each object to be split is recorded, and step S3.2 correspondences are pseudo-
Code 5-9 rows, the wherein row of false code the 7th represent and the mark point of each object to be split be put into list openlist, step
S3.3 correspondence false code 10-18 rows.
In the present embodiment, step S3.2 is for example：Object to be split in image includes foreground object and background object, prospect
The mark point of the object mark A with foreground object, the mark point mark B with background object of background object, then will
The correspondence position of the mark point of foreground object and the mark point of background object in distance matrix dist is assigned in undirected weighted graph
Be worth and be entered as 1 for the 0, correspondence position by the mark point of foreground object in undirected weighted graph in matrix of consequence label, will be undirected
Correspondence position of the mark point of background object in matrix of consequence label is entered as 2 in weighted graph.And then, step S3.3 is for example：
Correspondence positions of the unlabelled pixel x in distance matrix dist is entered as initial value infinity and in matrix of consequence label
In correspondence position be entered as 0, the correspondence position of the mark point y of foreground object in distance matrix dist is entered as 0 and in knot
Correspondence position in fruit matrix label is entered as 1；When step S3.3 is performed, update unlabelled using apart from renewal function
The assignment of correspondence positions of the pixel x in distance matrix dist, when utilization apart from renewal function (formula (2)) according to unmarked
Pixel x and mark point y between side right weight wx,yUnlabelled pixel x after the renewal being calculated is in distance matrix
Assignment dist (the x of the correspondence position in distnew) for 100 when (infinitely great less than its initial value), receive unmarked after the renewal
Correspondence positions of the pixel x in distance matrix dist assignment 100, and by unlabelled pixel x in matrix of consequence
The assignment of the correspondence position in label is updated to correspondence position assignment of the mark point y of foreground object in matrix of consequence label
It is 1, foreground object is belonging to which show unlabelled pixel x.
In the present embodiment, occur for the mistake for avoiding accumulative effect to cause as far as possible is divided, it is very big that a growth rate is set
Exponential type function increase influence of the gray scale difference to path cost between pixel.If by the second sensitive factor n in formula (1)
It is set to 1, it is assumed that the distance that pixel p is differed with pixel f is that pixel value difference is 1 between 3 location of pixels and each pixel,
And the distance of pixel p and pixel b for 1 location of pixels and pixel value differ be 2 when, mistake will be caused to divide：dist
Distance：e2＜ 3*e1.If the value of the second sensitive factor n is increased into 2, then the distance that mistake is divided can be increased to e2.So
And, if infinitely increasing by the second sensitive factor n to exchange the correct division ability for these slight changes for, probably make
Obtain the algorithm pixel somewhat violent for some changes and lose certain judgement, i.e., beyond operational precision scope (e30^2
=∞, when picture element interpolation is more than or equal to 30, will lose the ability of judging distance).So, in this implementation in renewal function
Example introduce forgetting factor θ, by existing dijkstra's algorithm apart from renewal function be changed into the present embodiment in distance update
Function：
dist(u′new)=dist (u ') * θ+wu,u′ (2)
If thoroughly to eliminate cumulative effect, then equation below is met apart from renewal function：
Dist (p → f) ＜ dist (p → b) (3)
Wherein, dist (p → f) represents the distance of pixel p to pixel f.If considering most extreme case, it is assumed that pixel
P to pixel f has infinite many pixels (c → ∞), and margin of image element is 1, then dist (p → f) is determined by equation below：
Dist (p → f)=θc-1*w+θc-2*w+…+w
Wherein w represents the unit side right weight in the figure determined by formula (1).Exceed in order to ensure occurring without result of calculation
Operational precision is limited, and the present embodiment specifies that the value of the second sensitive factor n in formula (1) is 1, then now w values are e.Finally
Dist (p → f) value such as formula (4) is described.Similarly, if the distance of pixel p and pixel b is 1, and grey scale pixel value is differed
2, then according to formula (1), dist (p → b) is e2
Forgetting factor can be obtained by formula (3) and formula (4)Therefore, to thoroughly eliminate tired
Product effect, then should set forgetting factor
Analyzed more than, the value of forgetting factor θ is set as 0.5 in the present embodiment.
In the present embodiment, the detailed process of step S4 is：
The nearest mark point of each unlabelled pixel in image is can obtain by step S3, with nearest mark point
Each unlabelled pixel of mark mark, so as to obtain the mark of each pixel in image, i.e. matrix of consequence label, will not
Generic boundary pixel is labeled as border to be split, i.e., different to adjacent position label value in matrix of consequence label
Region is marked, then classification boundaries are the curve that these mark points are linked to be, and completes image segmentation purpose.
When user is dissatisfied to classification results, user can increase mark point on the basis of existing classification results, return
Step S2 carries out further Optimum Classification, untill classification results are satisfied with.
As described above, the present invention is by being introduced into multi-source beeline thought in image segmentation field, and it is prominent a kind of
Efficient multi-source shortest path distance algorithm (finds the mark point closest with it of unlabelled pixel in step S3
Method) this problem is solved, reaching more fast and accurately image segmentation purpose.More specifically, the present invention is treated point first
Cut image and be weighted figure conversion, pixel difference is bigger between (relative) adjacent pixel, and weights are bigger.Afterwards, mark point is set
Be placed within object to be split, from traditional photoshop set mark point in object edge position to be split different, this hair
Bright labeling process more succinctly facilitates, and different from the labeling method for directly setting threshold value, and controllability of the invention is stronger.
In specific solution multi-source shortest path distance problem, different from traditional Floyd and Dijkstra, the present invention goes up at runtime
Keep consistent with dijkstra's algorithm also, different from the purpose of single source path planning of Dijkstra, the present invention can be in fortune
The determination work to the most short source point of all unmarked points in figure is completed in the case of going only once.Due to this kind of beeline
Introducing, compared to Random Walk Algorithm, the present invention can effectively overcome it to scheme due to balance background, caused by prospect gap
As the not good problem of details segmentation effect.The present invention in the case of seed point identical with Random Walk Algorithm setting, for thin
The division at section and edge is more.Also, due to the introducing of multi-source shortest path distance algorithm, the shadow of noise in image can be overcome
Ring.Importantly, due to the introducing of forgetting factor θ, the multi-source shortest path distance algorithm in the present invention is for remote locations
The pixel classifications of point are more accurate, and rational θ can be same not past Computing precision with the cooperation of n in formula (1)
When be prevented effectively from cumulative effect for remote pixel mistake divide.
In image segmentation operational efficiency, the present invention can realize more quickly scheming in the case of the minimum internal memory of consumption
As segmentation purpose.The experiment proved that, for the image of 480*640 resolution ratio, sliced time of the invention is only 0.2 second,
Cut 0.47 second of algorithm compared to figure and 1.0 seconds present invention of Random Walk Algorithm have embodied the advantage in efficiency.And
System committed memory aspect, the present invention cuts algorithm and Random Walk Algorithm still better than compared to figure.Specific experiment result such as table 1
It is shown.
Time and memory consumption of the algorithms of different of table 1 in image segmentation
Obviously, the above embodiment of the present invention is only intended to clearly illustrate example of the present invention, and is not right
The restriction of embodiments of the present invention, for those of ordinary skill in the field, may be used also on the basis of the above description
To make other changes in different forms, all of implementation method cannot be exhaustive here, it is every to belong to this hair
Obvious change that bright technical scheme is extended out changes row still in protection scope of the present invention.
Claims (8)
1. a kind of interactive image segmentation method based on multi-source shortest path distance, it is characterised in that the method includes as follows
Step：
Side right weight between S1, the pixel in the pixel value feature calculation image of image, and according between pixel
Image is converted into undirected weighted graph by side right again；
S2, interior of articles each to be split in the picture mark multiple pixels as the mark point of each object to be split respectively,
The mark point of object to be split carries the respective mark of object to be split；
S3, the mark point closest with it is found according to undirected weighted graph to all unlabelled pixels in image, so that
The problem of dividing the image into is converted into multi-source Shortest Path Searching problem；
S4, the mark each unlabelled pixel of mark with the mark point closest with each unlabelled pixel, obtain figure
The mark that each pixel is carried as in, and be partitioning boundary by the boundary marker of the pixel with different identification.
2. the interactive image segmentation method based on multi-source shortest path distance according to claim 1, it is characterised in that
Step S1 further includes following sub-step：
S1.1, the reachable domain that calculating side right weight is set；
The side right between the pixel in each pixel and the reachable domain centered on each pixel in S1.2, calculating image
Weight, computing formula is as follows：
In formula, wi,jIt is the side right weight between pixel i and pixel j；IiRepresent the gray scale of pixel i, δ be first it is sensitive because
Son, n is the second sensitive factor；
S1.3, image is converted into by undirected weighted graph according to the side right between pixel again.
3. the interactive image segmentation method based on multi-source shortest path distance according to claim 2, it is characterised in that
Up to domain be set to 5~20 close on it is reachable.
4. the interactive image segmentation method based on multi-source shortest path distance according to claim 2, it is characterised in that
IiIt is the rgb value of pixel, gray value or texture eigenvalue.
5. the interactive image segmentation method based on multi-source shortest path distance according to claim 2, it is characterised in that
Second sensitive factor n is set to 1, and the first sensitive factor δ is set to 1 if image is gray level image, if image is colour
Then the first sensitive factor δ is set to 9 to image.
6. the interactive image segmentation method based on multi-source shortest path distance according to claim 1, it is characterised in that
Mark point in step S2 is the continuous pixel of multiple in the multiple discontinuous pixel of mark or the continuous lines of mark
Point.
7. the interactive image segmentation method based on multi-source shortest path distance according to claim 1, it is characterised in that
Step S3 further includes following sub-step：
S3.1, the distance matrix that infinity is with the initial value of the size such as undirected weighted graph and each element is set up, and set up use
In deposit final segmentation result and be with the initial value of the size such as distance matrix and each element 0 matrix of consequence；
S3.2, the correspondence position by the mark point of each object to be split in undirected weighted graph in distance matrix be entered as 0, by nothing
The mark with each object to be split is entered as to correspondence position of the mark point of each object to be split in weighted graph in matrix of consequence
The corresponding numerical value of the respective mark of note point；
S3.3, the assignment using the correspondence position apart from renewal function each unlabelled pixel of renewal in distance matrix, when
Unlabelled pixel receives renewal in the assignment before the assignment of the correspondence position in distance matrix is less than its renewal after renewal
The assignment of correspondence position of the unlabelled pixel in distance matrix afterwards, and by the unlabelled pixel in matrix of consequence
The assignment of correspondence position be updated to correspondence of the mark point in matrix of consequence represented in the side right weight in renewal function
The assignment of position；
It is as follows apart from renewal function：
dist(u′new)=dist (u ') * θ+wu,u′
In formula, dist (u ') represents the assignment of correspondence positions of the unlabelled pixel u ' in distance matrix dist, and θ is something lost
Forget the factor, wu,u′Represent the side right weight between unlabelled pixel u ' and mark point u, dist (u 'new) represent update after not
The assignment of correspondence positions of the pixel u ' of mark in distance matrix dist.
8. the interactive image segmentation method based on multi-source shortest path distance according to claim 7, it is characterised in that
Forgetting factor
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611039325.6A CN106780506B (en) | 2016-11-21 | 2016-11-21 | Interactive image segmentation method based on multisource path shortest distance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611039325.6A CN106780506B (en) | 2016-11-21 | 2016-11-21 | Interactive image segmentation method based on multisource path shortest distance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106780506A true CN106780506A (en) | 2017-05-31 |
| CN106780506B CN106780506B (en) | 2020-05-08 |
Family
ID=58974932
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611039325.6A Expired - Fee Related CN106780506B (en) | 2016-11-21 | 2016-11-21 | Interactive image segmentation method based on multisource path shortest distance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106780506B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108614554A (en) * | 2018-05-03 | 2018-10-02 | 南京理工大学 | A kind of robot multi-source shortest path planning method based on region limitation |
| WO2020000960A1 (en) * | 2018-06-25 | 2020-01-02 | 北京达佳互联信息技术有限公司 | Image separation method, apparatus, computer device and storage medium |
| CN114396962A (en) * | 2022-02-25 | 2022-04-26 | 北京世纪高通科技有限公司 | Reachable domain generation method, device, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101714253A (en) * | 2009-12-04 | 2010-05-26 | 西安电子科技大学 | Interactive image segmentation correcting method based on geodesic active region models |
| CN102542543A (en) * | 2012-01-06 | 2012-07-04 | 西安电子科技大学 | Block similarity-based interactive image segmenting method |
| CN104036497A (en) * | 2014-05-26 | 2014-09-10 | 西安电子科技大学 | Graph cut interactive image segmentation algorithm based on local coefficient of variation |
| CN104063876A (en) * | 2014-01-10 | 2014-09-24 | 北京理工大学 | Interactive image segmentation method |
| CN104537355A (en) * | 2015-01-12 | 2015-04-22 | 中南大学 | Remarkable object detecting method utilizing image boundary information and area connectivity |
-
2016
- 2016-11-21 CN CN201611039325.6A patent/CN106780506B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101714253A (en) * | 2009-12-04 | 2010-05-26 | 西安电子科技大学 | Interactive image segmentation correcting method based on geodesic active region models |
| CN102542543A (en) * | 2012-01-06 | 2012-07-04 | 西安电子科技大学 | Block similarity-based interactive image segmenting method |
| CN104063876A (en) * | 2014-01-10 | 2014-09-24 | 北京理工大学 | Interactive image segmentation method |
| CN104036497A (en) * | 2014-05-26 | 2014-09-10 | 西安电子科技大学 | Graph cut interactive image segmentation algorithm based on local coefficient of variation |
| CN104537355A (en) * | 2015-01-12 | 2015-04-22 | 中南大学 | Remarkable object detecting method utilizing image boundary information and area connectivity |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108614554A (en) * | 2018-05-03 | 2018-10-02 | 南京理工大学 | A kind of robot multi-source shortest path planning method based on region limitation |
| WO2020000960A1 (en) * | 2018-06-25 | 2020-01-02 | 北京达佳互联信息技术有限公司 | Image separation method, apparatus, computer device and storage medium |
| US11494916B2 (en) | 2018-06-25 | 2022-11-08 | Beijing Dajia Internet Information Technology Co., Ltd. | Method for separating image and computer device |
| CN114396962A (en) * | 2022-02-25 | 2022-04-26 | 北京世纪高通科技有限公司 | Reachable domain generation method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106780506B (en) | 2020-05-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Philion et al. | Lift, splat, shoot: Encoding images from arbitrary camera rigs by implicitly unprojecting to 3d | |
| CN103577475B (en) | A kind of picture mechanized classification method, image processing method and its device | |
| Noma et al. | Interactive image segmentation by matching attributed relational graphs | |
| CN103207879B (en) | The generation method and apparatus of image index | |
| CN109613006A (en) | A kind of fabric defect detection method based on end-to-end neural network | |
| CN110334687A (en) | A kind of pedestrian retrieval Enhancement Method based on pedestrian detection, attribute study and pedestrian's identification | |
| CN110309888A (en) | A kind of image classification method and system based on layering multi-task learning | |
| CN106780506A (en) | A kind of interactive image segmentation method based on multi-source shortest path distance | |
| CN107330897A (en) | Image partition method and its system | |
| CN108229559A (en) | Dress ornament detection method, device, electronic equipment, program and medium | |
| CN109711266A (en) | Dress ornament attribute tags recognition methods based on neural network model sophisticated category | |
| CN109087330A (en) | It is a kind of based on by slightly to the moving target detecting method of smart image segmentation | |
| Wang et al. | Interactive multilabel image segmentation via robust multilayer graph constraints | |
| CN104778458A (en) | Textile pattern retrieval method based on textural features | |
| Nguyen et al. | Dynamic fuzzy clustering and its application in motion segmentation | |
| CN108170823A (en) | Hand-drawn interactive three-dimensional model retrieval method based on high-level semantic attribute understanding | |
| Craparotta et al. | A siamese neural network application for sales forecasting of new fashion products using heterogeneous data | |
| Szemenyei et al. | ROBO: robust, fully neural object detection for robot soccer | |
| CN102360432B (en) | Semantic marking method for image scene based on geodesic transmission | |
| Szilágyi et al. | Comparison of Various Improved-Partition Fuzzy c-Means Clustering Algorithms in Fast Color Reduction | |
| CN113378842A (en) | Recommendation method based on segmented image feature extraction | |
| CN109427068A (en) | Deep learning based on super-pixel automatic marking is divided into segmentation method | |
| CN102184524B (en) | Neighbourhood learning culture gene image segmentation method based on standard cut | |
| Kaya et al. | Fuzzy forecast combining for apparel demand forecasting | |
| CN116993983A (en) | Seedling family clothing pattern segmentation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200508 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |