CN103366379A - Level set medical image segmentation method based on heredity kernel fuzzy clustering - Google Patents

Level set medical image segmentation method based on heredity kernel fuzzy clustering Download PDF

Info

Publication number
CN103366379A
CN103366379A CN2013103219552A CN201310321955A CN103366379A CN 103366379 A CN103366379 A CN 103366379A CN 2013103219552 A CN2013103219552 A CN 2013103219552A CN 201310321955 A CN201310321955 A CN 201310321955A CN 103366379 A CN103366379 A CN 103366379A
Authority
CN
China
Prior art keywords
cluster
colony
medical image
function
image
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
CN2013103219552A
Other languages
Chinese (zh)
Other versions
CN103366379B (en
Inventor
朱家明
张天平
盛朗
居小平
王涛
高飞
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.)
SINOWAYS MEDICAL TECHNOLOGY Co Ltd
Yangzhou University
Original Assignee
SINOWAYS MEDICAL TECHNOLOGY Co Ltd
Yangzhou University
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 SINOWAYS MEDICAL TECHNOLOGY Co Ltd, Yangzhou University filed Critical SINOWAYS MEDICAL TECHNOLOGY Co Ltd
Priority to CN201310321955.2A priority Critical patent/CN103366379B/en
Publication of CN103366379A publication Critical patent/CN103366379A/en
Application granted granted Critical
Publication of CN103366379B publication Critical patent/CN103366379B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention provides a level set medical image segmentation method based on heredity kernel fuzzy clustering and relates to the application of medical image segmentation. According to the level set medical image segmentation method disclosed by the invention, a heredity kernel fuzzy clustering algorithm is utilized to obtain an optimal clustering result of a medical image to be treated and then the clustering result is applied to an initial outline of an LBF (Local Binary Fitting) model to carry out the segmentation on the image, so that a blood vessel image has high segmentation efficiency and accuracy.

Description

Level set medical image partition method based on hereditary Kernel fuzzy clustering
Technical field
The invention belongs to image processing field, relate to the application of medical image segmentation.Be specifically related to genetic algorithm (GA), Fuzzy c-means clustering algorithm (KFCM) and Level Set Method are in the application in image segmentation field.
Background technology
Medical image segmentation is subject to the attention of international academic community as an important component part in the image segmentation field always, and a large amount of scholars obtain remarkable achievement in this field.But for same clinical image, because different application purposes, interested tissue will be different.So how selecting according to demand suitable partitioning algorithm is a so far insoluble difficult problem of medical image segmentation field.Such as, the MR image for same width of cloth brain can be divided into according to different demands: the extraction of the extraction of brain tissue, the classification of brain tissue, privileged site brain tissue structure etc., the dividing method between this three differs greatly.Therefore, also there is not so far a kind of unified dividing method effectively to cut apart dissimilar medical images.
Nearly decades, Level Set Method has obtained considerable house show in the image segmentation field, and the method was proposed by Osher and Sethian at first in 1988.The method is converted in the high one-dimensional space the implicit mode of level set function curve evolvement with the closed curve evolution problem and finds the solution, it can process the problem that image topology changes preferably, thereby so that the topological structure of segmentation result and initial curve is irrelevant, thereby be widely used in the image segmentation field.
At present, using more widely based on the image Segmentation Technology of level set, model has: peripheral type parted pattern, regional parted pattern, shape prior type parted pattern and mixed type parted pattern.The peripheral type parted pattern mainly utilizes the gradient information of image to stop curve evolvement, namely promotes the evolution curve by local edge information and approaches objective contour.Because the peripheral type model has only utilized the marginal information of image, area information for image is not then used, so it is not good for this model segmentation effect of the weak-edge image that does not have obvious graded, in order to address this problem, the domain type parted pattern is suggested, this model utilizes the area information of image to promote the evolution curve and approaches to objective contour, most typical domain type model is the C-V Level Set Models that han and Vese propose on the basis of Level Set Method, and this model utilization evolution inside or outside of curve section gray average promotes curve evolvement.Shape prior pattern type utilizes level set function to express the shape of sample, directly the develop principal ingredient of curve of level set function is analyzed, then by the training to target shape obtain target prior shape, then add the shape constraining item in movable contour model, the difficult point of this model is to express accurately objective trait.The mixed type parted pattern mainly is that three is before interosculated, and uses part and the global information of image.
Although having obtained immense success based on level set image segmentation area research personnel, still have problems.Although the C-V model has been used the area information of image, it can't obtain satisfied segmentation result to the inhomogeneous image of intensity profile.For this reason, Li and his team have proposed local two-value match (LBF) model, and this model defines local two-value match amount by introducing kernel function.The LBF model can effectively be cut apart the inhomogeneous image of intensity profile, and higher precision is arranged, but, because this model has adopted the gradient information of image border, so it is relatively responsive to choosing of initial profile, and choosing difference and may cause the far from each other of segmentation result along with initial profile.Its antimierophonic performance is not high simultaneously, for baroque medical image segmentation DeGrain.
Summary of the invention
The object of the invention is to the weak point for above-mentioned LBF model, a kind of medical image cutting method based on hereditary Kernel fuzzy clustering is proposed, with solve the LBF model for initial profile the responsive and low problem of noise robustness, thereby the raising medical image cut apart quality.
Technical solution of the present invention is: the gray-scale value that will obtain first the dendrogram picture of medical image carries out cluster, again with each cluster centre
Figure 772829DEST_PATH_IMAGE001
Be encoded into character string with 8 bits, form cluster colony, calculate respectively the objective function of cluster
Figure 405936DEST_PATH_IMAGE002
With the ideal adaptation degree
Figure 647561DEST_PATH_IMAGE003
, to the colony's service condition when former generation
Figure 188264DEST_PATH_IMAGE004
, condition is satisfied then to finish algorithm, and obtain Optimal cluster centers, and calculate optimum membership function by following formula,
Figure 2013103219552100002DEST_PATH_IMAGE001
Utilize maximum membership grade principle
Figure 635743DEST_PATH_IMAGE006
Determine the classification that medical image gradation data collection is affiliated, obtain the optimum cluster image
Figure 99085DEST_PATH_IMAGE007
Pass through again Determine the initial profile of Level Set Models;
At last, being applied to the LBF model cuts apart medical image.
The present invention utilizes hereditary Kernel fuzzy clustering algorithm to obtain the optimum cluster result of pending medical image, and the initial profile that then cluster result is applied to the LBF model is to Image Segmentation Using, can accomplish blood-vessel image is had higher efficient and the accuracy cut apart.
Concrete grammar may further comprise the steps:
1) obtain the dendrogram picture of medical image:
If former space sample
Figure 941456DEST_PATH_IMAGE009
, can be obtained to a feature space by the nuclear Nonlinear Mapping:
Figure 2013103219552100002DEST_PATH_IMAGE003
The dot product in so former sample space at feature space can be with the Mercer kernel representation:
Figure 2013103219552100002DEST_PATH_IMAGE004
Kernel function is got gaussian kernel function
Figure 2013103219552100002DEST_PATH_IMAGE005
1-1) establish initialization cluster centre number
Figure 334949DEST_PATH_IMAGE013
, and according to cluster numbers the gray-scale value of medical image is carried out cluster, with each cluster centre
Figure 11918DEST_PATH_IMAGE014
Be encoded into character string with 8 bits, form cluster colony;
1-2) establish the maximum evolutionary generation of genetic algorithm , group size
Figure 136049DEST_PATH_IMAGE016
, crossover probability
Figure 811881DEST_PATH_IMAGE017
, the variation probability
Figure 925331DEST_PATH_IMAGE018
, nuclear parameter
Figure 850561DEST_PATH_IMAGE019
, the weight index
Figure 74869DEST_PATH_IMAGE020
, generate at random Individual initial population
Figure 155138DEST_PATH_IMAGE021
, and order
Figure 302085DEST_PATH_IMAGE022
1-3) will The string of binary characters decoding of individual colony represents with real number form, utilizes following formula to calculate respectively the objective function of cluster With the ideal adaptation degree
Figure 435760DEST_PATH_IMAGE024
, wherein
Figure 335583DEST_PATH_IMAGE025
Be
Figure 901694DEST_PATH_IMAGE026
The cluster centre of class,
Figure 406624DEST_PATH_IMAGE027
Be the picture of this center at corresponding nuclear space:
Figure 2013103219552100002DEST_PATH_IMAGE006
With
And have
1-4) to the colony's service condition when former generation
Figure 524119DEST_PATH_IMAGE030
, finish algorithm when satisfying condition, and skip next step, directly carry out step 1-6), and obtain Optimal cluster centers, calculate optimum membership function by following formula; If do not satisfy
Figure 149136DEST_PATH_IMAGE030
Then carry out next step 1-5):
Figure 2013103219552100002DEST_PATH_IMAGE008
1-5) when not reaching maximum evolutionary generation
Figure 555026DEST_PATH_IMAGE015
The time, order
Figure 462939DEST_PATH_IMAGE032
,Utilize roulette dish method, single-point bracketing method, mutation operator to current colony select, intersection, mutation operation, produce colony of new generation
Figure 942462DEST_PATH_IMAGE021
, simultaneously reservation is worked as the large individuality of former generation fitness and is arrived
Figure 910418DEST_PATH_IMAGE021
In, turn back to step 1-3) cluster again;
Utilize genetic operator to colony select, crossover and mutation operation, produce colony of new generation ,
Wherein select the operator design as follows:
A) select operation: the individuality that is genetic to a new generation according to the cumulative selection of selected probability; Individual selected probability is:
Figure 2013103219552100002DEST_PATH_IMAGE009
Wherein Certain individuality in the expression colony,
Figure 456620DEST_PATH_IMAGE035
Be the size of colony,
Figure 774469DEST_PATH_IMAGE024
Be the ideal adaptation degree;
B) interlace operation: in the character string of coding, by crossover probability a point of crossing is set, then exchanges the position of two pairs of character strings of back;
C) mutation operation: the probability that variation is set is 5%, selects at random a son touring, and is inserted into a position at random;
Utilize maximum membership grade principle
Figure 257141DEST_PATH_IMAGE036
Determine the classification that medical image gradation data collection is affiliated, obtain the optimum cluster image
Figure 242414DEST_PATH_IMAGE037
2) pass through
Figure 552173DEST_PATH_IMAGE038
Determine the initial profile of Level Set Models;
3) initial profile that obtains according to step 2 is applied to the LBF model medical image is cut apart;
If given image I:
Figure 357318DEST_PATH_IMAGE039
Be given original image territory, d is the image dimension; With curve C as the zero level set function , the zone
Figure 719346DEST_PATH_IMAGE041
With
Figure 200006DEST_PATH_IMAGE042
Corresponding respectively
Figure 226868DEST_PATH_IMAGE043
With
Figure 2013103219552100002DEST_PATH_IMAGE010
, the LBF model can be expressed as:
Figure 2013103219552100002DEST_PATH_IMAGE011
Wherein
Figure 429813DEST_PATH_IMAGE046
Weight coefficient,
Figure 209550DEST_PATH_IMAGE047
Be respectively the zone
Figure 73601DEST_PATH_IMAGE048
With
Figure 887973DEST_PATH_IMAGE049
Being similar to of gradation of image, and evolution rule is:
Figure 977469DEST_PATH_IMAGE051
Gaussian kernel function, the Heaviside function, namely
Figure 645211DEST_PATH_IMAGE052
In addition, add the length regularization term
Figure 314089DEST_PATH_IMAGE053
Guarantee the slickness of contour curve, in order to guarantee after a period of time not loss of gloss slip of horizontal odd function, increase the energy penalty term simultaneously:
Adopt gradient descent flow and the variational method to obtain following curve evolvement equation:
Figure 2013103219552100002DEST_PATH_IMAGE014
Wherein
Figure 533215DEST_PATH_IMAGE056
The Dirac function,
Figure 56601DEST_PATH_IMAGE057
With
Figure 220866DEST_PATH_IMAGE058
Be defined as:
Figure DEST_PATH_IMAGE015
To EVOLUTION EQUATION Discretize obtains following expression:
Figure DEST_PATH_IMAGE016
Try to achieve new level set function according to following formula
Figure 450673DEST_PATH_IMAGE062
, by
Figure 914015DEST_PATH_IMAGE062
positive and negatively obtain new cut zone, and whether the determined level set function restrain, if not then forward step 2 to and continue iteration with new level set function and cut zone, otherwise stops iteration, obtains final segmentation result.
Description of drawings
Fig. 1 is the blood vessel original image.
Fig. 2 is initial profile 1 figure of LBF model.
Fig. 3 is that the LBF model is for the segmentation result figure of initial profile 1.
Fig. 4 is initial profile 2 figure of LBF model.
Fig. 5 is that the LBF model is for the segmentation result figure of initial profile 2.
Fig. 6 is segmentation result figure of the present invention.
Fig. 7 is brain MR original image.
Fig. 8 is for adding the brain MR image of noise.
Fig. 9 is the segmentation result figure of LBF model.
Figure 10 is segmentation result figure of the present invention.
Embodiment
Effect of the present invention further specifies by following emulation experiment:
Experimental situation is Matlab7.1, CORE i3 CPU, internal memory 4GB.Experimental data is angiographic image and MR image.
The specific implementation process is as follows:
Step 1: the dendrogram picture that obtains medical image.
Suppose former space sample
Figure 727250DEST_PATH_IMAGE063
, can be obtained to a feature space by the nuclear Nonlinear Mapping:
Figure DEST_PATH_IMAGE017
The dot product in so former sample space at feature space can be with the Mercer kernel representation:
This paper kernel function is got gaussian kernel function
Figure DEST_PATH_IMAGE018
1-1 sets initialization cluster centre number
Figure 531816DEST_PATH_IMAGE067
, and according to cluster numbers the gray-scale value of medical image is carried out cluster, with each cluster centre
Figure 884300DEST_PATH_IMAGE025
Be encoded into character string with 8 bits, form cluster colony.
1-2 sets the maximum evolutionary generation of genetic algorithm
Figure 826848DEST_PATH_IMAGE015
, group size
Figure 733624DEST_PATH_IMAGE016
, crossover probability
Figure 419821DEST_PATH_IMAGE017
, the variation probability , nuclear parameter
Figure 740260DEST_PATH_IMAGE068
, the weight index
Figure 665491DEST_PATH_IMAGE020
, generate at random
Figure 358641DEST_PATH_IMAGE016
Individual initial population , and order
Figure 704488DEST_PATH_IMAGE069
1-3 will
Figure 585857DEST_PATH_IMAGE016
The string of binary characters decoding of individual colony represents with real number form, utilizes
Figure DEST_PATH_IMAGE019
With
Figure 795438DEST_PATH_IMAGE071
Calculate respectively the objective function of cluster
Figure 250690DEST_PATH_IMAGE023
With the ideal adaptation degree , wherein
Figure 919886DEST_PATH_IMAGE025
Be The cluster centre of class,
Figure 847708DEST_PATH_IMAGE072
Be the picture of this center at corresponding nuclear space, and have
Figure DEST_PATH_IMAGE020
Figure DEST_PATH_IMAGE021
1-4 is to the colony's service condition when former generation
Figure 964065DEST_PATH_IMAGE075
, condition is satisfied then to finish algorithm, obtains Optimal cluster centers, and passes through
Calculate optimum membership function, and go to (6), otherwise go to (5).
1-5 ought not reach maximum evolutionary generation
Figure 838798DEST_PATH_IMAGE015
The time, order
Figure 746711DEST_PATH_IMAGE077
, utilize roulette dish method, single-point bracketing method, mutation operator to current colony select, intersection, mutation operation, produce colony of new generation
Figure 757392DEST_PATH_IMAGE021
, simultaneously reservation is worked as the large individuality of former generation fitness and is arrived
Figure 725348DEST_PATH_IMAGE021
In, go to (3).
Utilize genetic operator to colony select, crossover and mutation operation, produce colony of new generation
Figure 290322DEST_PATH_IMAGE021
,
Genetic operator is and how selects, crossover and mutation.Wherein select the operator design as follows:
(1) selects operation
Adopt the roulette wheel selection method to select the higher individuality of fitness and be genetic to the next generation, namely be genetic to the individuality of a new generation according to the cumulative selection of selected probability.Individual selected probability is:
Figure 1926DEST_PATH_IMAGE078
Wherein
Figure 867114DEST_PATH_IMAGE034
Certain individuality in the expression colony,
Figure 5971DEST_PATH_IMAGE035
Be the size of colony,
Figure 556776DEST_PATH_IMAGE024
Be the ideal adaptation degree.
(2) interlace operation
Interlace operation adopts the single-point bracketing method that two pairing individualities are operated, and namely by crossover probability a point of crossing is set in the character string of coding, then exchanges the position of two pairs of character strings of back.
(3) mutation operation
The probability that variation is set is 5%, selects at random a son touring, and is inserted into a position at random.
1-6 utilizes maximum membership grade principle
Figure 72071DEST_PATH_IMAGE079
Determine the classification that medical image gradation data collection is affiliated, obtain the optimum cluster image
Figure 791765DEST_PATH_IMAGE037
Step 2: by
Figure 101524DEST_PATH_IMAGE080
Determine the initial profile of Level Set Models.
Step 3: according to the initial profile that step 2 obtains, be applied to the LBF model medical image is cut apart.
Suppose given image I:
Figure 641089DEST_PATH_IMAGE081
Be given original image territory, d is the image dimension.With curve C as the zero level set function
Figure 694496DEST_PATH_IMAGE082
, the zone
Figure 534276DEST_PATH_IMAGE083
With
Figure 14936DEST_PATH_IMAGE084
Corresponding respectively
Figure 41798DEST_PATH_IMAGE085
With
Figure 633316DEST_PATH_IMAGE086
, the LBF model can be expressed as:
Figure DEST_PATH_IMAGE023
Wherein ,
Figure 493322DEST_PATH_IMAGE089
Weight coefficient,
Figure 888531DEST_PATH_IMAGE090
,
Figure 171745DEST_PATH_IMAGE091
Be respectively the zone
Figure 994207DEST_PATH_IMAGE085
With
Figure 261241DEST_PATH_IMAGE086
Being similar to of gradation of image, and evolution rule is:
Figure DEST_PATH_IMAGE024
Gaussian kernel function, the Heaviside function, namely
In addition, add the length regularization term
Figure 345554DEST_PATH_IMAGE095
Guarantee the slickness of contour curve, in order to guarantee after a period of time not loss of gloss slip of horizontal odd function, increase the energy penalty term simultaneously:
Adopt gradient descent flow and the variational method to obtain following curve evolvement equation:
Figure DEST_PATH_IMAGE026
Wherein
Figure 504637DEST_PATH_IMAGE098
The Dirac function,
Figure 215104DEST_PATH_IMAGE057
With
Figure 490228DEST_PATH_IMAGE058
Be defined as:
Figure DEST_PATH_IMAGE027
。
To EVOLUTION EQUATION
Figure 436242DEST_PATH_IMAGE100
Discretize obtains following expression:
Figure DEST_PATH_IMAGE028
Try to achieve new level set function according to following formula
Figure 712819DEST_PATH_IMAGE102
, by
Figure 679638DEST_PATH_IMAGE102
positive and negatively obtain new cut zone, and whether the determined level set function restrain, if not then forward step 2 to and continue iteration with new level set function and cut zone, otherwise stops iteration, obtains final segmentation result.
Experiment 1. comparisons for blood-vessel image this paper model and LBF model:
From blood vessel original image Fig. 1, carry out the LBF model intercepting of initial profile 1 as shown in Figure 2.The LBF model is for the segmentation result of initial profile 1 as seen from Figure 3, and Fig. 5 is the segmentation result to the initial profile 2 of the LBF model of Fig. 4.
It is a width of cloth that the present invention adopts above experimental data
Figure 388968DEST_PATH_IMAGE103
The blood vessel gray level image, can obtain the cluster result of hereditary Kernel fuzzy clustering algorithm according to step 1.Cluster centre is in step 1
Figure 339607DEST_PATH_IMAGE104
, greatest iteration number
Figure 690954DEST_PATH_IMAGE105
, window size N is
Figure 43438DEST_PATH_IMAGE106
, choose at random the initial population number , crossover probability
Figure 892762DEST_PATH_IMAGE108
, the variation probability
Figure 313379DEST_PATH_IMAGE109
, nuclear parameter
Figure 254790DEST_PATH_IMAGE110
, the weight index
Figure 368240DEST_PATH_IMAGE111
After setting, initial parameter calculates respectively the objective function of cluster according to step 1-3
Figure 293470DEST_PATH_IMAGE023
With the ideal adaptation degree
Figure 517778DEST_PATH_IMAGE024
Judge according to step 1-4 for existing colony, wherein
Figure 558371DEST_PATH_IMAGE112
, when satisfying
Figure 311563DEST_PATH_IMAGE113
The time forward step 1-6 to, finally obtain optimum cluster, when not satisfying above-mentioned condition, go to step 1-5, and do not reach maximum evolutionary generation
Figure 724090DEST_PATH_IMAGE015
The time, order , utilize roulette dish method, single-point bracketing method, mutation operator to current colony select, intersection, mutation operation, produce colony of new generation
Figure 402513DEST_PATH_IMAGE021
, simultaneously reservation is worked as the large individuality of former generation fitness and is arrived
Figure 592186DEST_PATH_IMAGE021
In, go to (3).Finally can obtain the optimum cluster result of blood-vessel image according to above step
Figure 492009DEST_PATH_IMAGE037
Then pass through Determine the initial profile of Level Set Models, herein
Figure 828629DEST_PATH_IMAGE115
It is enlightening carat function.Final step is the initial profile substitution level set function that utilizes hereditary kernel clustering to obtain, and carries out EVOLUTIONARY COMPUTATION according to step 3, time step in this step
Figure 189203DEST_PATH_IMAGE116
, the Heaviside function , finally obtain this paper segmentation result Fig. 6.
By experiment, and comparison diagram 6 and Fig. 5 can find out, the segmentation result of the Different L BF model of choosing along with initial profile is far from each other, and segmentation error rate is higher, and is relatively more responsive for choosing of initial profile.And therefore this paper model does not exist the initial profile sensitive issue owing to do not rely on initial profile, and can find out that from experimental result this paper model can accomplish blood-vessel image is had higher efficient and the accuracy cut apart.
Experiment 2. comparisons for MR image this paper model and LBF model.
Experimental data is to add Gaussian noise The MR image.Fig. 8 is 0 for add average in Fig. 7, and variance is 0.02 MR image.Cluster centre is in step 1
Figure 305561DEST_PATH_IMAGE104
, greatest iteration number
Figure 102616DEST_PATH_IMAGE119
, window size N is
Figure 180293DEST_PATH_IMAGE106
, choose at random the initial population number , crossover probability
Figure 364467DEST_PATH_IMAGE121
, the variation probability
Figure 332423DEST_PATH_IMAGE122
, nuclear parameter , the weight index
Figure 609000DEST_PATH_IMAGE111
After setting, initial parameter calculates respectively the objective function of cluster according to step 1-3
Figure 474188DEST_PATH_IMAGE023
With the ideal adaptation degree
Figure 81887DEST_PATH_IMAGE024
Judge according to step 1-4 for existing colony, wherein
Figure 665315DEST_PATH_IMAGE124
, when satisfying
Figure 915031DEST_PATH_IMAGE125
The time forward step 1-6 to, finally obtain optimum cluster, when not satisfying above-mentioned condition, go to step 1-5, and do not reach maximum evolutionary generation
Figure 900304DEST_PATH_IMAGE015
The time, order , utilize roulette dish method, single-point bracketing method, mutation operator to current colony select, intersection, mutation operation, produce colony of new generation
Figure 248164DEST_PATH_IMAGE021
, simultaneously reservation is worked as the large individuality of former generation fitness and is arrived
Figure 301571DEST_PATH_IMAGE021
In, go to (3).Finally can obtain the optimum cluster result of blood-vessel image according to above step
Figure 141351DEST_PATH_IMAGE037
Then pass through Determine the initial profile of Level Set Models, herein
Figure 383293DEST_PATH_IMAGE115
It is enlightening carat function.Final step is the initial profile substitution level set function that utilizes hereditary kernel clustering to obtain, and carries out EVOLUTIONARY COMPUTATION according to step 3, time step in this step
Figure 240391DEST_PATH_IMAGE127
, the Heaviside function
Figure DEST_PATH_IMAGE128
, finally obtain this paper segmentation result Figure 10.
And Fig. 9 is the segmentation result figure of LBF model.
By experiment, comparison diagram 9,10 can find out: the LBF model is responsive for noise ratio, and this paper invention algorithm LBF model of comparing has better segmentation precision and efficient, has higher robustness.Wherein the LBF model average computation time is 2.25s, and this paper invention required time is 0.58s, has improved to a great extent counting yield.

Claims (2)

1. based on the level set medical image partition method of hereditary Kernel fuzzy clustering, its feature spy is: the gray-scale value that will obtain first the dendrogram picture of medical image carries out cluster, again with each cluster centre
Figure 208628DEST_PATH_IMAGE001
Be encoded into character string with 8 bits, form cluster colony, calculate respectively the objective function of cluster
Figure 390211DEST_PATH_IMAGE002
With the ideal adaptation degree , to the colony's service condition when former generation
Figure 2013103219552100001DEST_PATH_IMAGE001
, condition is satisfied then to finish algorithm, and obtain Optimal cluster centers, and calculate optimum membership function by following formula,
Figure 2013103219552100001DEST_PATH_IMAGE002
Utilize maximum membership grade principle
Figure 878644DEST_PATH_IMAGE006
Determine the classification that medical image gradation data collection is affiliated, obtain the optimum cluster image
Figure 504797DEST_PATH_IMAGE007
Pass through again
Figure 157495DEST_PATH_IMAGE008
Determine the initial profile of Level Set Models;
At last, being applied to the LBF model cuts apart medical image.
2. described level set medical image partition method based on hereditary Kernel fuzzy clustering according to claim 1, its feature spy is may further comprise the steps:
1) obtain the dendrogram picture of medical image:
If former space sample
Figure 527297DEST_PATH_IMAGE009
, can be obtained to a feature space by the nuclear Nonlinear Mapping:
Figure 683472DEST_PATH_IMAGE010
The dot product in so former sample space at feature space can be with the Mercer kernel representation:
Figure 11685DEST_PATH_IMAGE011
Kernel function is got gaussian kernel function
Figure 2013103219552100001DEST_PATH_IMAGE003
1-1) establish initialization cluster centre number
Figure 528434DEST_PATH_IMAGE013
, and according to cluster numbers the gray-scale value of medical image is carried out cluster, with each cluster centre
Figure 70274DEST_PATH_IMAGE001
Be encoded into character string with 8 bits, form cluster colony;
1-2) establish the maximum evolutionary generation of genetic algorithm
Figure 38230DEST_PATH_IMAGE014
, group size
Figure 134362DEST_PATH_IMAGE015
, crossover probability
Figure 845966DEST_PATH_IMAGE016
, the variation probability
Figure 242312DEST_PATH_IMAGE017
, nuclear parameter
, the weight index
Figure 964597DEST_PATH_IMAGE019
, generate at random Individual initial population , and order
Figure 40504DEST_PATH_IMAGE021
1-3) will
Figure 111228DEST_PATH_IMAGE015
The string of binary characters decoding of individual colony represents with real number form, utilizes following formula to calculate respectively the objective function of cluster
Figure 695793DEST_PATH_IMAGE002
With the ideal adaptation degree
Figure 535573DEST_PATH_IMAGE003
, wherein
Figure 16233DEST_PATH_IMAGE001
Be
Figure 105412DEST_PATH_IMAGE022
The cluster centre of class,
Figure 696930DEST_PATH_IMAGE023
For this center at corresponding nuclear space
Picture:
Figure 2013103219552100001DEST_PATH_IMAGE004
With
And have
Figure 2013103219552100001DEST_PATH_IMAGE005
1-4) to the colony's service condition when former generation , finish algorithm when satisfying condition, and skip next step, directly carry out step 1-6), and obtain Optimal cluster centers, calculate optimum membership function by following formula; If do not satisfy
Figure DEST_PATH_IMAGE006
Then carry out next step 1-5);
Figure DEST_PATH_IMAGE007
1-5) when not reaching maximum evolutionary generation
Figure 651296DEST_PATH_IMAGE014
The time, order
Figure 449488DEST_PATH_IMAGE028
, utilize roulette dish method, single-point bracketing method, mutation operator to current colony select, intersection, mutation operation, produce colony of new generation
Figure 648388DEST_PATH_IMAGE020
, simultaneously reservation is worked as the large individuality of former generation fitness and is arrived
Figure 317267DEST_PATH_IMAGE020
In, turn back to step 1-3) cluster again;
Utilize genetic operator to colony select, crossover and mutation operation, produce colony of new generation
Figure 310631DEST_PATH_IMAGE020
,
Wherein select the operator design as follows:
A) select operation: the individuality that is genetic to a new generation according to the cumulative selection of selected probability; Individual selected probability is:
Figure DEST_PATH_IMAGE008
Wherein Certain individuality in the expression colony is the size of colony, is the ideal adaptation degree;
B) interlace operation: in the character string of coding, by crossover probability a point of crossing is set, then exchanges the position of two pairs of character strings of back;
C) mutation operation: the probability that variation is set is 5%, selects at random a son touring, and is inserted into a position at random;
1-6) utilize maximum membership grade principle to determine the classification that medical image gradation data collection is affiliated, obtain the optimum cluster image;
2) pass through
Figure 977739DEST_PATH_IMAGE033
Determine the initial profile of Level Set Models;
3) initial profile that obtains according to step 2 is applied to the LBF model medical image is cut apart;
If given image I: ,
Figure 572985DEST_PATH_IMAGE035
Be given original image territory, d is the image dimension; With curve C as the zero level set function , the zone
Figure 149777DEST_PATH_IMAGE037
With
Figure 452582DEST_PATH_IMAGE038
Corresponding respectively
Figure 403221DEST_PATH_IMAGE039
With
Figure 285726DEST_PATH_IMAGE040
, the LBF model can be expressed as:
Figure DEST_PATH_IMAGE009
Wherein
Figure 111917DEST_PATH_IMAGE042
,
Figure 549851DEST_PATH_IMAGE043
Weight coefficient,
Figure 970468DEST_PATH_IMAGE044
With
Figure 443038DEST_PATH_IMAGE045
Be respectively the zone
Figure 87646DEST_PATH_IMAGE046
With Being similar to of gradation of image, and evolution rule is:
Figure 298681DEST_PATH_IMAGE049
Gaussian kernel function, the Heaviside function, namely
Figure DEST_PATH_IMAGE011
In addition, add the length regularization term
Figure 526717DEST_PATH_IMAGE051
Guarantee the slickness of contour curve, in order to guarantee after a period of time not loss of gloss slip of horizontal odd function, increase the energy penalty term simultaneously:
Figure 554716DEST_PATH_IMAGE052
Adopt gradient descent flow and the variational method to obtain following curve evolvement equation:
Figure DEST_PATH_IMAGE012
Wherein
Figure 988289DEST_PATH_IMAGE054
The Dirac function,
Figure 419270DEST_PATH_IMAGE055
With
Figure 985381DEST_PATH_IMAGE056
Be defined as:
Figure DEST_PATH_IMAGE013
To EVOLUTION EQUATION
Figure 647623DEST_PATH_IMAGE058
Discretize obtains following expression:
Figure DEST_PATH_IMAGE014
Try to achieve new level set function according to following formula
Figure 935702DEST_PATH_IMAGE060
, by
Figure 826298DEST_PATH_IMAGE060
positive and negatively obtain new cut zone, and whether the determined level set function restrain, if not then forward step 2 to and continue iteration with new level set function and cut zone, otherwise stops iteration, obtains final segmentation result.
CN201310321955.2A 2013-07-29 2013-07-29 Level set medical image partition method based on heredity Kernel fuzzy clustering Active CN103366379B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310321955.2A CN103366379B (en) 2013-07-29 2013-07-29 Level set medical image partition method based on heredity Kernel fuzzy clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310321955.2A CN103366379B (en) 2013-07-29 2013-07-29 Level set medical image partition method based on heredity Kernel fuzzy clustering

Publications (2)

Publication Number Publication Date
CN103366379A true CN103366379A (en) 2013-10-23
CN103366379B CN103366379B (en) 2016-08-10

Family

ID=49367647

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310321955.2A Active CN103366379B (en) 2013-07-29 2013-07-29 Level set medical image partition method based on heredity Kernel fuzzy clustering

Country Status (1)

Country Link
CN (1) CN103366379B (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103942799A (en) * 2014-04-25 2014-07-23 哈尔滨医科大学 Breast ultrasounography image segmentation method and system
CN104616308A (en) * 2015-02-12 2015-05-13 大连民族学院 Multiscale level set image segmenting method based on kernel fuzzy clustering
CN106127734A (en) * 2016-06-13 2016-11-16 西安电子科技大学 MRI image dividing method based on obscure idea and level set framework
CN106504239A (en) * 2016-10-25 2017-03-15 南通大学 A kind of method of liver area in extraction ultrasonoscopy
CN107567637A (en) * 2015-04-30 2018-01-09 皇家飞利浦有限公司 Brain tissue is classified
CN108182684A (en) * 2017-12-22 2018-06-19 河南师范大学 A kind of image partition method and its device based on weighting kernel fuzzy cluster
CN110648340A (en) * 2019-09-29 2020-01-03 惠州学院 Method and device for processing image based on binary system and level set
CN111027546A (en) * 2019-12-05 2020-04-17 北京嘉楠捷思信息技术有限公司 Character segmentation method and device and computer readable storage medium
CN112686916A (en) * 2020-12-28 2021-04-20 淮阴工学院 Curved surface reconstruction system based on heterogeneous multi-region CT scanning data processing
CN112907526A (en) * 2021-02-07 2021-06-04 电子科技大学 LBF-based satellite telescope lens surface defect detection method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101719277A (en) * 2009-12-31 2010-06-02 华中科技大学 Method for partitioning genetic fuzzy clustering image
CN101980298A (en) * 2010-11-04 2011-02-23 西安电子科技大学 Multi-agent genetic clustering algorithm-based image segmentation method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101719277A (en) * 2009-12-31 2010-06-02 华中科技大学 Method for partitioning genetic fuzzy clustering image
CN101980298A (en) * 2010-11-04 2011-02-23 西安电子科技大学 Multi-agent genetic clustering algorithm-based image segmentation method

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103942799B (en) * 2014-04-25 2017-02-01 哈尔滨医科大学 Breast ultrasounography image segmentation method and system
CN103942799A (en) * 2014-04-25 2014-07-23 哈尔滨医科大学 Breast ultrasounography image segmentation method and system
CN104616308A (en) * 2015-02-12 2015-05-13 大连民族学院 Multiscale level set image segmenting method based on kernel fuzzy clustering
CN107567637B (en) * 2015-04-30 2022-01-25 皇家飞利浦有限公司 Brain tissue classification
CN107567637A (en) * 2015-04-30 2018-01-09 皇家飞利浦有限公司 Brain tissue is classified
CN106127734B (en) * 2016-06-13 2019-01-11 西安电子科技大学 MRI image dividing method based on obscure idea and level set frame
CN106127734A (en) * 2016-06-13 2016-11-16 西安电子科技大学 MRI image dividing method based on obscure idea and level set framework
CN106504239B (en) * 2016-10-25 2019-06-21 南通大学 A kind of method of liver area in extraction ultrasound image
CN106504239A (en) * 2016-10-25 2017-03-15 南通大学 A kind of method of liver area in extraction ultrasonoscopy
CN108182684A (en) * 2017-12-22 2018-06-19 河南师范大学 A kind of image partition method and its device based on weighting kernel fuzzy cluster
CN108182684B (en) * 2017-12-22 2021-06-25 河南师范大学 Image segmentation method and device based on weighted kernel function fuzzy clustering
CN110648340A (en) * 2019-09-29 2020-01-03 惠州学院 Method and device for processing image based on binary system and level set
CN110648340B (en) * 2019-09-29 2023-03-17 惠州学院 Method and device for processing image based on binary system and level set
CN111027546A (en) * 2019-12-05 2020-04-17 北京嘉楠捷思信息技术有限公司 Character segmentation method and device and computer readable storage medium
CN111027546B (en) * 2019-12-05 2024-03-26 嘉楠明芯(北京)科技有限公司 Character segmentation method, device and computer readable storage medium
CN112686916A (en) * 2020-12-28 2021-04-20 淮阴工学院 Curved surface reconstruction system based on heterogeneous multi-region CT scanning data processing
CN112686916B (en) * 2020-12-28 2024-04-05 淮阴工学院 Curved surface reconstruction system based on heterogeneous multi-region CT scanning data processing
CN112907526A (en) * 2021-02-07 2021-06-04 电子科技大学 LBF-based satellite telescope lens surface defect detection method
CN112907526B (en) * 2021-02-07 2022-04-19 电子科技大学 LBF-based satellite telescope lens surface defect detection method

Also Published As

Publication number Publication date
CN103366379B (en) 2016-08-10

Similar Documents

Publication Publication Date Title
CN103366379A (en) Level set medical image segmentation method based on heredity kernel fuzzy clustering
CN109145939B (en) Semantic segmentation method for small-target sensitive dual-channel convolutional neural network
CN105512289B (en) Image search method based on deep learning and Hash
CN107506761A (en) Brain image dividing method and system based on notable inquiry learning convolutional neural networks
CN106529574B (en) Image classification method based on sparse autocoder and support vector machines
CN107169504B (en) A kind of hand-written character recognition method based on extension Non-linear Kernel residual error network
CN109523463A (en) A kind of face aging method generating confrontation network based on condition
CN105608692B (en) Polarization SAR image segmentation method based on deconvolution network and sparse classification
CN104537673B (en) Infrared Image Segmentation based on multi thresholds and adaptive fuzzy clustering
CN101625755A (en) Image division method based on watershed-quantum evolution clustering algorithm
CN106650767A (en) Flood forecasting method based on cluster analysis and real time correction
CN109902564A (en) A kind of accident detection method based on the sparse autoencoder network of structural similarity
CN103593674B (en) A kind of cervical lymph node ultrasonoscopy feature selection method
CN103679136B (en) Hand back vein identity recognition method based on combination of local macroscopic features and microscopic features
CN108664838A (en) Based on the monitoring scene pedestrian detection method end to end for improving RPN depth networks
CN101419706A (en) Jersey wear flokkit and balling up grading method based on image analysis
CN107679368A (en) PET/CT high dimensional feature level systems of selection based on genetic algorithm and varied precision rough set
CN108509843A (en) A kind of face identification method of the Huber constraint sparse codings based on weighting
CN106683102A (en) SAR image segmentation method based on ridgelet filters and convolution structure model
CN103020979B (en) Image segmentation method based on sparse genetic clustering
CN105427313A (en) Deconvolutional network and adaptive inference network based SAR image segmentation method
CN109948776A (en) A kind of confrontation network model picture tag generation method based on LBP
CN106611423A (en) SAR image segmentation method based on ridge wave filter and deconvolution structural model
CN105117707A (en) Regional image-based facial expression recognition method
CN108681689A (en) Based on the frame per second enhancing gait recognition method and device for generating confrontation network

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