CN109035268A - A kind of self-adaptive projection method method - Google Patents
A kind of self-adaptive projection method method Download PDFInfo
- Publication number
- CN109035268A CN109035268A CN201810719050.3A CN201810719050A CN109035268A CN 109035268 A CN109035268 A CN 109035268A CN 201810719050 A CN201810719050 A CN 201810719050A CN 109035268 A CN109035268 A CN 109035268A
- Authority
- CN
- China
- Prior art keywords
- curve
- energy
- model
- function
- self
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/149—Segmentation; Edge detection involving deformable models, e.g. active contour models
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A kind of self-adaptive projection method method, including, the original image pixels gray value of input is adjusted to 0-255 range by S1;S2 chooses the initial active contour line of gray scale adjustment image;S3, the relative entropy of calculated curve interior zone and perimeter;S4 constructs self-adaptive projection method movable contour model, and the interior zone of contour curve and the relative entropy of perimeter are introduced into movable contour model replaces λ respectively2And λ1As its energy function internal energy and external energy weight coefficient, the automatic adjusument of weight coefficient is realized, complete image segmentation.The present invention can enable curve that can adaptively carry out parameter selection in evolutionary process, image segmentation is accurate with the internal energy and external energy weight coefficient of automatic adjusument movable contour model energy function.
Description
Technical field
The present invention relates to Medical Image Segmentation Techniques fields, more particularly to a kind of self-adaptive projection method method.
Background technique
Image segmentation is exactly to be extracted the region of interest in image using features such as the gray scale of image, texture or shapes
Come, is committed step one of of the image procossing to image analysis, for computer vision, pattern-recognition, Medical Image Processing etc.
It has very important significance.In image procossing, the superiority and inferiority for dividing quality determines subsequent image analysis such as feature extraction, mesh
The quality of the mark tasks such as not.In recent years, movable contour model due to its calculate efficiently, be suitable for modeling and its can will be to be processed
The advantages that priori knowledge of problem is together with various image processing algorithm effective integrations are regarded in field of image processing and computer
It is greatly developed and is widely applied in feel field.Particularly, one of the main stream approach for having become image segmentation, has obtained state
The extensive concern of inside and outside scholar, especially can be in conjunction with the geometric active contour model based on region of image information.
Image segmentation is the basis of image procossing and computer vision neighborhood, and segmentation result directly affects having for follow-up work
Effect property and efficiency.In recent years, all images be can be suitably used for without any one algorithm in numerous image partition methods.Image segmentation
Method is generally based on image, i.e., it is right in turn information to be obtained from image using image gradient, brightness or texture etc.
Image is split, and mainly has clustering procedure, region growing, figure to cut, active contour scheduling algorithm.Movable contour model (Active
Contour Model, ACM) since expression formula is simple, computational efficiency is high, the deformation suitable for modeling or extracting arbitrary shape
Profile has had more successfully application in Image Edge-Detection, image segmentation and motion tracking.The algorithm is being schemed first
A closed curve is initialized as in, then by minimizing energy functional, moves to initial curve on object boundary, or obtain
Image segmentation result.Movable contour model can be divided into parametric active contour model and geometric active according to curve representation form difference
Skeleton pattern.
Active contour is explicitly described as a parametrization energy curve by parametric active contour model, in control wheel profile
Work of the internal energy (internal force) and attraction contour line of bending and stretching to the mobile external energy (external force) in expectation target edge
With lower deformation, object boundary is finally converged to.Snake model is the parametric active contour model being suggested at first, is thought substantially
Some control points to constitute certain shapes are want as template (contour line), by the elastic deformation of template itself, with image local
Feature, which matches, reaches reconciliation, i.e. certain energy function minimization, completes the segmentation to image;The purpose is to reconcile upper layer knowledge
With this contradiction of low layer pictures feature.Mainly there are three parts to form for the realization process of Snake model, is initial profile line first
Determination can measure the selection of curve;Second is the calculating of energy, including the calculating of internal energy and external energy;Third
It is the minimum point i.e. Numerical Implementation process how these nodes find energy.The advantages of Snake model: first is that passing through scale sky
Between, minimization energy, expands capture region, while reducing computation complexity, realizes objective contour from thick to thin
Quick and precisely divide;Second is that the visual problems such as edge, line, objective contour are all handled by unified mechanism;Third is that can be with
It instructs profile to develop in conjunction with more high layer informations, is appropriate for user's interaction.
The theoretical basis of geometric active contour model is curve evolution theory and level set thought.The general idea of the model
It is the zero level collection that planar closed curve is impliedly expressed as to higher-dimension toroidal function, by minimizing energy functional, by curve
EVOLUTION EQUATION be converted into the partial differential equation of higher-dimension curved surface level set function, be then iterated evolution, make zero level transporting something containerized
It moves to going on objective contour.Geodesic curve geometric active contour model (geodesic active contours, GAC) is a kind of allusion quotation
The model based on boundary segmentation of type identifies the boundary of object using image gradient come the evolution of driving curve.The model proposes
New way based on the objective contour detection that minimal path calculates, by the parametric active contour model that is minimized based on energy with
Geometric active contour model based on curve evolvement theory connects.Chan-Vese model, i.e. CV model are classical bases
Geometric active contour model in region.The model is based on Mumford-Shah functional, based on image by average gray phase
The biggish homogeneous region of difference forms this it is assumed that be divided using the difference of the average gray between target and background
It cuts.CV model instructs curve evolvement using image overall grayscale information, and computation complexity is low, and curve can in evolutionary process
It is automatic to carry out topologies change, it can preferably handle weak-edge image.Energy (local binary is fitted based on local binary
Fitting energy) regional activity skeleton pattern, i.e. LBF model and a kind of classical model.The model is CV model
Global two-value fitting energy functional is changed to be fitted energy definition by part (Variable Area) two-value of kernel function of Gaussian function
Energy functional, by pixel grey scale local variance information inside and outside calculated curve come driving curve evolution.LBF model extraction part ash
The objective contour in the uneven image of gray scale can be obtained using level set movements by spending information, and without weight in iterative process
New initialization level set function, it is only necessary to calculate a partial differential equation (partial differential equation,
PDE).The model greatly simplifies operand, has the advantages that calculating is simple, curve change in topology can be effectively treated, be suitable for segmentation
The images such as nonuniformity region, weak boundary and vessel-like structure.Topography is fitted (local image fitting, LIF) energy
Amount model is that have the model being widely recognized as after CV, LBF.The model foundation energy letter of local fit image
Number, can be regarded as fitting image and original image difference constraint condition, and use gaussian filtering after iteration each time just
Then change level set function, calculating process is without carrying out reinitializing for level set.Compared with LBF model, LIF model is can
While obtaining similar accuracy rate, operation efficiency is greatly improved.In addition to above-mentioned several typical models, learned there are also very much
Person from different perspectives studies geometric active contour model, no longer repeats one by one.
Snake model is very big to the selection dependence of initial curve, due to the characteristic distributions of external force, the position of initial curve
Directly influence final profile line it is accurate whether;Model is unable to the change in topology of accommodation curve, automatically realizes division and close
And because the curve that parameter of curve can only define a closure can not be by curve segmentation at two, it is difficult to converge to spill side
Boundary, therefore the application range of Snake model is limited to a certain extent;GAC model is the gradient information by object boundary
Image segmentation is carried out, therefore, good segmentation result can be obtained when image to be split is strong edge, but work as image to be split
When for weak edge, curve is easily propagated through in weak edge, causes accurately to divide;CV model utilizes image overall grayscale information
Curve evolvement is instructed, computation complexity is low, and curve can carry out in evolutionary process topologies change automatically, can be preferable
Weak-edge image is handled, but CV model cannot preferably handle the medicine figure for being widely present heterogeneity and structural complexity
Picture;Gaussian kernel function is introduced data fit term by LBF model, is driven by pixel grey scale local variance information inside and outside calculated curve
Moving curve develops, and the segmentation problem of the uneven image of gray scale cannot be handled by preferably solving CV model, but the model is to initial
There is certain sensibility in the position of profile, while poor to strong noise image robustness.LIF model is based on area information
The improvement of LBF establishes new topography's fitting energy function, reduces computational complexity, but the difference meeting of initial profile position
Certain influence is generated on segmentation effect.The above movable contour model, either CV model or LBF model all have more ginseng
Number, and to parameter sensitivity, the unreasonable segmentation result that will lead to mistake of parameter setting, therefore split-run test will be to parameter every time
It is adjusted, especially the weight parameter λ of the internal energy and external energy of driving curve evolution1And λ2, this is undoubtedly to method
Realization causes certain obstruction.
Therefore, in view of the shortcomings of the prior art, providing a kind of self-adaptive projection method method to overcome the deficiencies of the prior art very
For necessity.
Summary of the invention
A kind of self-adaptive projection method method is provided it is an object of the invention to avoid the deficiencies in the prior art place, it can
With the internal energy and external energy weight coefficient of automatic adjusument movable contour model energy function, enable curve in evolutionary process
In can adaptively carry out parameter selection, move in or out evolution curve adaptively according to image information, so as to more
Add accurate convergent object boundary, realizes image segmentation.
Above-mentioned purpose of the invention is realized by following technological means.
A kind of self-adaptive projection method method is provided, is carried out as follows:
Original image pixels gray value is adjusted to 0-255 range as input picture by S1;
S2 chooses the initial active contour line of input picture;
The relative entropy D of S3, calculated curve interior zone and perimeterRE(p1||p2) and DRE(p2||p1);
S4 constructs self-adaptive projection method movable contour model, by the interior zone of the S3 contour curve being calculated and
The relative entropy D of perimeterRE(p1||p2) and DRE(p2||p1) be introduced into movable contour model respectively instead of λ2And λ1As its energy
Flow function internal energy and external energy weight coefficient realize the automatic adjusument of weight coefficient, complete image segmentation.
Preferably, in above-mentioned steps S3, especially by the opposite of such as under type calculated curve interior zone and perimeter
Entropy DRE(p1||p2) and DRE(p2||p1):
p1(x|φ),p2(x | φ) is the probability density letter of evolution curve C interior zone and perimeter in image respectively
Number, then:
RE is a kind of directive measurement, is not symmetry value, i.e. DRE(p1||p2)≠DRE(p2||p1), therefore can generation respectively
For λ2And λ1.Relative entropy can change with the evolution of contour curve, therefore can realize the automatic adjusument of weight coefficient.By formula
It is apparent from, works as p1(x | φ) > p2(x | φ) when, DRE(p1||p2) > DRE(p2||p1), work as p1(x | φ) < p2(x | φ) when, DRE(p1
||p2) < DRE(p2||p1)。
Preferably, above-mentioned steps S4, the weight coefficient of internal energy and external energy in geometric active contour model
It is replaced, is carried out adaptively selected with relative entropy.Common geometric active contour model has CV model, LBF model etc..The present invention with
For selecting LBF model, by the relative entropy D of the interior zone of the S3 contour curve being calculated and perimeterRE(p1||p2)
And DRE(p2||p1) be introduced into movable contour model respectively instead of λ2And λ1It is weighed as its energy function internal energy and external energy
Weight coefficient, constructs a self-adaptive projection method model --- ALBF.
ALBF model, it is known that I: Ω → R is image to be split, whereinIt is the domain of image, closed curve C is
The zero level collection of level set function φ, i.e. C=z | and φ (z)=0 }, by the energy function of the ALBF model of level set expression are as follows:
Wherein, μ, ν >=0 are the weight coefficients for being respectively regularization term and length penalty term;DRE(p1||p2),DRE(p2||
p1) it is the relative entropy that can change with curve evolvement introduced;f1It is the Gaussian function K for being σ through varianceσThe song obtained after filtering
The mean value of neighborhood, f inside line2It is the Gaussian function K for being σ through varianceσThe mean value of the curved exterior neighborhood obtained after filtering, δ (φ)
It is Dirac function and Heaviside function respectively with H (φ);
Equation first item F (φ, f1,f2) it is the data item that local binary is fitted energy, it calculates evolution curve local neighborhood
Internal energy and external energy, driving curve develop;Equation Section 2 P (φ) is level set regularization term, makes level set letter
Number φ remains symbolic measurement in evolutionary process, without reinitializing;Equation Section 3 L (φ) is to zero level wheel
The punishment of wide length, to keep contour line C smooth;
Fixed φ, minimization energy function formula (3) obtain f1(x) and f2(x) expression formula are as follows:
Finally, obtaining the evolution about zero level collection using the calculus of variations and gradient descent method minimization energy function formula (5)
Equation:
Wherein, e1=∫ΩKσ(x-y)|I(y)-f1(x)|2Dy, e2=∫ΩKσ(x-y)|I(y)-f2(x)|2dy。
The smooth function H that Heaviside function H is defined by the formula in equation (4)εApproximation obtains:
Dirac function is then is defined as:
Preferably, the original image pixels gray value of input is adjusted to 0- particular by linear transformation by above-mentioned steps S1
255 ranges.
Self-adaptive projection method method of the invention carries out as follows: S1, by original image pixels gray value tune
The whole 0-255 range that arrives is as input picture;S2 chooses the initial active contour line of input picture;S3, calculated curve interior zone
With the relative entropy D of perimeterRE(p1||p2) and DRE(p2|| p1);S4 constructs self-adaptive projection method movable contour model, will
The interior zone for the contour curve that S3 is calculated and the relative entropy D of perimeterRE(p1||p2) and DRE(p2|| p1) introducing activity
λ is replaced in skeleton pattern respectively2And λ1As its energy function internal energy and external energy weight coefficient, weight coefficient is realized
Automatic adjusument, complete image segmentation.Self-adaptive projection method method of the invention, can be with automatic adjusument active contour mould
The internal energy and external energy weight coefficient of type energy function, enable curve that can adaptively carry out parameter choosing in evolutionary process
Select, move in or out evolution curve adaptively according to image information, so as to more accurately convergent object boundary,
Realize image segmentation.
Detailed description of the invention
Using attached drawing, the present invention is further illustrated, but the content in attached drawing is not constituted to any limit of the invention
System.
Fig. 1 is the flow chart of self-adaptive projection method method of the present invention.
Fig. 2 is LBF model local neighborhood schematic diagram.
Fig. 3 is the split-run test comparative result figure for being directed to composograph in different ways.
Fig. 4 is the segmentation result comparison diagram for being directed to blood-vessel image in different ways.
Fig. 5 is to carry out white matter split-run test comparative result figure for Brain MR Image in different ways.
Fig. 6 is the comparative result figure being split in different ways for kidney CT image.
Specific embodiment
The invention will be further described with the following Examples.
Embodiment 1.
A kind of self-adaptive projection method method, as shown in Figure 1, carrying out as follows:
Original image pixels gray value is adjusted to 0-255 range as input picture by S1.It can will be former by linear transformation
Beginning image pixel gray level value is adjusted to 0-255 range.
S2 chooses the initial active contour line of input picture.
The relative entropy D of S3, calculated curve interior zone and perimeterRE(p1||p2) and DRE(p2||p1), especially by such as
The relative entropy D of under type calculated curve interior zone and perimeterRE(p1||p2) and DRE(p2||p1):
p1(x|φ),p2(x | φ) is the probability density letter of evolution curve C interior zone and perimeter in image respectively
Number, then:
When similitude is smaller between two distributions, DREValue is bigger, otherwise smaller.It should be noted that RE is that one kind has
The measurement in direction is not symmetry value, i.e. DRE(p1||p2)≠DRE(p2||p1), therefore λ can be replaced respectively2And λ1.Work as curvilinear inner
When homogeney is greater than curved exterior homogeney, i.e. p1(x | φ) > p2(x | φ), it is shared in data item that external homogeney can be increased
Specific gravity reduces its influence to curve evolvement.When similarly curved exterior homogeney is greater than internal homogeney, increase internal homogeneity
Property proportion.It is apparent from by formula, works as p1(x | φ) > p2(x | φ) when, DRE(p1||p2) > DRE(p2||p1), work as p1(x|φ)
< p2(x | φ) when, DRE(p1||p2) < DRE(p2||p1)。
S4 constructs self-adaptive projection method movable contour model, by the interior zone of the S3 contour curve being calculated and
The relative entropy D of perimeterRE(p1|| p2) and DRE(p2||p1) be introduced into movable contour model respectively instead of λ2And λ1As its energy
Flow function internal energy and external energy weight coefficient realize the automatic adjusument of weight coefficient, complete image segmentation.
In fact, the weight coefficient of internal energy and external energy can be with relative entropy come generation in geometric active contour model
It replaces, carries out adaptively selected.Because present invention experiment is by taking LBF as an example, self-adaptive projection method model may be simply referred to as ALBF model.
LBF model is selected, by the relative entropy D of the interior zone of the S3 contour curve being calculated and perimeterRE(p1||
p2) and DRE(p2|| p1) be introduced into movable contour model respectively instead of λ2And λ1As its energy function internal energy and external energy
Measure weight coefficient;
ALBF model, it is known that I: Ω → R is image to be split, whereinIt is the domain of image, closed curve C is
The zero level collection of level set function φ, i.e. C=z | and φ (z)=0 }, by the energy function of the ALBF model of level set expression are as follows:
Wherein, μ, ν >=0 are the weight coefficients for being respectively regularization term and length penalty term;DRE(p1|| p2),DRE(p2||
p1) it is the relative entropy that can change with curve evolvement introduced;f1It is the Gaussian function K for being σ through varianceσThe song obtained after filtering
The mean value of neighborhood, f inside line2It is the Gaussian function K for being σ through varianceσThe mean value of the curved exterior neighborhood obtained after filtering, δ (φ)
It is Dirac function and Heaviside function respectively with H (φ);
Equation first item F (φ, f1,f2) it is the data item that local binary is fitted energy, it calculates evolution curve local neighborhood
Internal energy and external energy, driving curve develop;Equation Section 2 P (φ) is level set regularization term, makes level set letter
Number φ remains symbolic measurement in evolutionary process, without reinitializing;Equation Section 3 L (φ) is to zero level wheel
The punishment of wide length, to keep contour line C smooth.When curve C develops to object boundary, energy function E (φ) obtains minimum
Value.Fig. 2 is LBF model local neighborhood, and by taking evolution curve local neighborhood interior zone as an example, curve C is contour curve, round wires institute
Envelope surface accumulates NxIt is the local neighborhood centered on x that size is determined by kernel function variances sigma, y is a bit in x neighborhood.If point y is in neighbour
Domain NxIt is interior in contour curve, point y distance center point x is remoter, and the energy is smaller;If y is in neighborhood NxOutside inherent contour curve,
The energy is zero;If point y is in neighborhood NxOutside, which is also zero.
Fixed φ, minimization energy function formula (3) obtain f1(x) and f2(x) expression formula are as follows:
Finally, obtaining the evolution about zero level collection using the calculus of variations and gradient descent method minimization energy function formula (5)
Equation:
Wherein, e1=∫ΩKσ(x-y)|I(y)-f1(x)|2Dy, e2=∫ΩKσ(x-y)|I(y)-f2(x)|2dy。
The smooth function H that Heaviside function H is defined by the formula in equation (4)εApproximation obtains:
Dirac function is then is defined as:
Self-adaptive projection method method of the invention is to introduce relative entropy Relative Entropy, RE) replace castor
λ in wide model1And λ2As the weight coefficient of internal energy and external energy, the method for automatic adjusument weight parameter.This hair
It is bright that model is improved for this cumbersome disadvantage of movable contour model parameter setting, calculated curve interior zone and outside area
The weighting parameter λ of internal energy and external energy that the RE in domain replaces driving curve to develop1And λ2.The introducing of RE enables algorithm in song
Can automatically adjust inside or outside of curve portion region proportion in line evolutionary process, i.e., evolution curve according to image information adaptively to
It is interior or be displaced outwardly, so that curve is preferably converged to object boundary, accurate completion image segmentation.The present invention is not necessarily to weight
Parameter is arranged one by one manually, and the introducing of relative entropy enables algorithm that can automatically adjust inside or outside of curve portion region institute during curve evolvement
Accounting weight, adaptive realization dynamic state of parameters are adjusted, and realize accurate image segmentation.
Embodiment 2.
Below in conjunction with specific test example, the invention will be further described.
Split-run test is carried out to a large amount of composograph and medical image, by new model ALBF and CV model and LBF mould
The segmentation effect of type compares.Table 1 is model parameter setting in each experiment, and Δ t is time step.
The setting of 1. experiment parameter of table
We have carried out split-run test to composograph first, and experimental result is as shown in Figure 3.It can be seen that from segmentation result,
The segmentation effect for the non-uniform image that CV model synthesizes three kinds is unsatisfactory;LBF algorithm is final under current parameter setting
Although contour line is higher than the completeness of CV model, fail to identify object boundary completely;ALBF algorithm is because can be adaptive
Selection inside and outside region proportion, the automatic parameter selection that carries out obtains optimal segmentation result.
Fig. 4 is the comparative result figure to real blood vessels image segmentation.The results show that CV model cannot correctly divide both
Blood-vessel image, although LBF model has correctly divided the second width blood-vessel image, treatment of details is not so good as ALBF model, in figure
The circled of arrow meaning.This is because the optimal inside and outside region weight coefficient of selection that ALBF algorithm is adaptive, makes final
Contour curve more approaching to reality object boundary.
White matter split-run test has been carried out to Brain MR Image, has as a result seen Fig. 5.In the segmentation result figure of CV and LBF algorithm, arrow
Erroneous segmentation all has occurred at head meaning etc., and ALBF algorithm can accurately split white matter of brain.It further illustrates
The superiority of ALBF algorithm auto-adaptive parameter selection.
A large amount of kidney CT images are divided in experiment, and image size is 512 × 512, and segmentation result is with expert's hand
Dynamic segmentation result is judgment basis, and experimental result is as shown in Figure 6.It can be seen that from segmentation result, when image contains more non-homogeneous
When region, CV model cannot be preferably split;LBF model algorithm detects edge dependent on gradient, detects object boundary
While erroneous segmentation easily occurs at the biggish edge of change of gradient;ALBF algorithm is maintained due to introducing RE to LBF model
The original advantage of LBF model, auto-adaptive parameter selection make final profile converge to object boundary well.
As it can be seen that the present invention improves model for this cumbersome disadvantage of movable contour model parameter setting, calculate bent
The weighting parameter λ of internal energy and external energy that line interior zone and the RE of perimeter replace driving curve to develop1And λ2。
The introducing of RE enables algorithm that can automatically adjust inside or outside of curve portion region proportion during curve evolvement, i.e., evolution curve according to
Image information adaptively moves in or out, and curve is made preferably to converge to object boundary, accurate completion image point
It cuts.
Finally it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention rather than protects to the present invention
The limitation of range, although the invention is described in detail with reference to the preferred embodiments, those skilled in the art should be managed
Solution, can with modification or equivalent replacement of the technical solution of the present invention are made, without departing from technical solution of the present invention essence and
Range.
Claims (6)
1. a kind of self-adaptive projection method method, which is characterized in that carry out as follows:
Original image pixels gray value is adjusted to 0-255 range as input picture by S1;
S2 chooses the initial active contour line of input picture;
The relative entropy D of S3, calculated curve interior zone and perimeterRE(p1||p2) and DRE(p2||p1);
S4 constructs self-adaptive projection method movable contour model, by the interior zone of the S3 contour curve being calculated and outside
The relative entropy D in regionRE(p1||p2) and DRE(p2||p1) be introduced into movable contour model respectively instead of λ2And λ1As its energy letter
Number internal energy and external energy weight coefficient realize the automatic adjusument of weight coefficient, complete image segmentation.
2. self-adaptive projection method method according to claim 1, which is characterized in that
In step S3, especially by the relative entropy D of such as under type calculated curve interior zone and perimeterRE(p1||p2) and DRE
(p2||p1):
p1(x | φ), p2(x | φ) is the probability density function of evolution curve C interior zone and perimeter in image respectively, then:
RE is a kind of directive measurement, is not symmetry value, i.e. DRE(p1||p2)≠DRE(p2||p1), therefore for replacing λ respectively2
And λ1;Relative entropy can change the automatic adjusument to realize weight coefficient with the evolution of contour curve;Work as p1(x | φ) > p2
(x | φ) when, DRE(p1||p2) > DRE(p2||p1), work as p1(x | φ) < p2(x | φ) when, DRE(p1||p2) < DRE(p2||p1)。
3. self-adaptive projection method method according to claim 2, which is characterized in that
The weight coefficient of internal energy and external energy carries out certainly by being replaced with relative entropy in geometric active contour model
Adapt to selection.
4. self-adaptive projection method method according to claim 3, which is characterized in that geometric active contour model is CV mould
Type.
5. self-adaptive projection method method according to claim 3, which is characterized in that
In step s 4, geometric active contour model selects LBF model, by the interior zone of the S3 contour curve being calculated and
The relative entropy D of perimeterRE(p1||p2) and DRE(p2||p1) be introduced into LBF model respectively instead of λ2And λ1As its energy letter
Number internal energy and external energy weight coefficient, construct a self-adaptive projection method model --- ALBF;
ALBF model, it is known that I: Ω → R is image to be split, whereinIt is the domain of image, closed curve C is level
The zero level collection of set function φ, i.e. C=z | and φ (z)=0 }, by the energy function of the ALBF model of level set expression are as follows:
Wherein, μ, ν >=0 are the weight coefficients for being respectively regularization term and length penalty term;DRE(p1||p2),DRE(p2||p1) be
The relative entropy that can change with curve evolvement introduced;f1It is the Gaussian function K for being σ through varianceσThe curvilinear inner obtained after filtering
The mean value of neighborhood, f2It is the Gaussian function K for being σ through varianceσThe mean value of the curved exterior neighborhood obtained after filtering, δ (φ) and H
(φ) is Dirac function and Heaviside function respectively;
Equation first item F (φ, f1,f2) it is the data item that local binary is fitted energy, it calculates the interior of evolution curve local neighborhood
Portion's energy and external energy, driving curve develop;Equation Section 2 P (φ) is level set regularization term, makes level set function φ
Symbolic measurement is remained in evolutionary process, without reinitializing;Equation Section 3 L (φ) is long to zero level profile
The punishment of degree, to keep contour line C smooth;
Fixed φ, minimization energy function formula (3) obtain f1(x) and f2(x) expression formula are as follows:
Finally, obtaining the evolution side about zero level collection using the calculus of variations and gradient descent method minimization energy function formula (5)
Journey:
Wherein, e1=∫ΩKσ(x-y)|I(y)-f1(x)|2Dy, e2=∫ΩKσ(x-y)|I(y)-f2(x)|2dy;
The smooth function H that Heaviside function H is defined by the formula in equation (4)εApproximation obtains:
Dirac function is then is defined as:
6. self-adaptive projection method method according to claim 1, it is characterised in that:
The original image pixels gray value of input is adjusted to 0-255 range particular by linear transformation by step S1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810719050.3A CN109035268A (en) | 2018-07-03 | 2018-07-03 | A kind of self-adaptive projection method method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810719050.3A CN109035268A (en) | 2018-07-03 | 2018-07-03 | A kind of self-adaptive projection method method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109035268A true CN109035268A (en) | 2018-12-18 |
Family
ID=65521579
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810719050.3A Pending CN109035268A (en) | 2018-07-03 | 2018-07-03 | A kind of self-adaptive projection method method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109035268A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685816A (en) * | 2018-12-27 | 2019-04-26 | 上海联影医疗科技有限公司 | Image segmentation method, device, equipment and storage medium |
CN109840914A (en) * | 2019-02-28 | 2019-06-04 | 华南理工大学 | A kind of Texture Segmentation Methods based on user's interactive mode |
CN110084824A (en) * | 2019-04-26 | 2019-08-02 | 山东财经大学 | Tongue body image partition method, system, equipment and medium based on symmetrical level set |
CN111784716A (en) * | 2020-06-04 | 2020-10-16 | 华中科技大学 | Sequence diagram image segmentation method and system based on ultrasonic CT |
-
2018
- 2018-07-03 CN CN201810719050.3A patent/CN109035268A/en active Pending
Non-Patent Citations (1)
Title |
---|
刘燕杰: "基于活动轮廓模型的医学图像分割", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685816A (en) * | 2018-12-27 | 2019-04-26 | 上海联影医疗科技有限公司 | Image segmentation method, device, equipment and storage medium |
CN109685816B (en) * | 2018-12-27 | 2022-05-13 | 上海联影医疗科技股份有限公司 | Image segmentation method, device, equipment and storage medium |
CN109840914A (en) * | 2019-02-28 | 2019-06-04 | 华南理工大学 | A kind of Texture Segmentation Methods based on user's interactive mode |
CN109840914B (en) * | 2019-02-28 | 2022-12-16 | 华南理工大学 | Texture segmentation method based on user interaction |
CN110084824A (en) * | 2019-04-26 | 2019-08-02 | 山东财经大学 | Tongue body image partition method, system, equipment and medium based on symmetrical level set |
CN110084824B (en) * | 2019-04-26 | 2020-03-27 | 山东财经大学 | Tongue image segmentation method, system, device and medium based on symmetric level set |
CN111784716A (en) * | 2020-06-04 | 2020-10-16 | 华中科技大学 | Sequence diagram image segmentation method and system based on ultrasonic CT |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Xian et al. | Automatic breast ultrasound image segmentation: A survey | |
Yin et al. | Large scale remote sensing image segmentation based on fuzzy region competition and Gaussian mixture model | |
CN110475505B (en) | Automatic segmentation using full convolution network | |
CN109035268A (en) | A kind of self-adaptive projection method method | |
Chakraborty et al. | Deformable boundary finding influenced by region homogeneity | |
CN107146228B (en) | A kind of super voxel generation method of brain magnetic resonance image based on priori knowledge | |
CN108932536A (en) | Human face posture method for reconstructing based on deep neural network | |
CN107004268A (en) | The automatic region of interest regional partition and registration of the Dynamic constrasted enhancement image of colorectal carcinoma | |
US7680335B2 (en) | Prior-constrained mean shift analysis | |
Kapse et al. | Literature survey on detection of brain tumor from MRI images | |
Deeparani et al. | Efficient image segmentation and implementation of K-means clustering | |
CN112419343A (en) | System and method for image segmentation | |
Chin et al. | Lips Contour Detection and Tracking Using Watershed Region-Based Active Contour Model and Modified $ H_ {\infty} $ | |
Sridevi et al. | Self-organizing neural networks for image segmentation based on multiphase active contour | |
Ghalib et al. | Artificial neural network based detection of renal tumors using CT scan image processing | |
Zanaty | An adaptive fuzzy C-means algorithm for improving MRI segmentation | |
CN116935133A (en) | Cardiovascular disease classification method and system based on SPECT image recognition | |
CN107729863A (en) | Human body refers to vein identification method | |
Ghosh et al. | Robust simultaneous registration and segmentation with sparse error reconstruction | |
Xiao et al. | Fast level set image and video segmentation using new evolution indicator operators | |
CN107492101B (en) | Multi-modal nasopharyngeal tumor segmentation algorithm based on self-adaptive constructed optimal graph | |
Ding et al. | Segmentation algorithm of medical exercise rehabilitation image based on HFCNN and IoT | |
Pal et al. | A brief survey on image segmentation based on quantum inspired neural network | |
Yuan et al. | Active contours driven by local intensity and local gradient fitting energies | |
Wang et al. | A robust statistics driven volume-scalable active contour for segmenting anatomical structures in volumetric medical images with complex conditions |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181218 |