CN105139398A - Multi-feature-based gray uneven image fast segmentation method - Google Patents
Multi-feature-based gray uneven image fast segmentation method Download PDFInfo
- Publication number
- CN105139398A CN105139398A CN201510531170.7A CN201510531170A CN105139398A CN 105139398 A CN105139398 A CN 105139398A CN 201510531170 A CN201510531170 A CN 201510531170A CN 105139398 A CN105139398 A CN 105139398A
- Authority
- CN
- China
- Prior art keywords
- function
- phi
- dtri
- iteration
- model
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
- G06T2207/10061—Microscopic image from scanning electron microscope
-
- 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
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30008—Bone
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30024—Cell structures in vitro; Tissue sections in vitro
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a multi-feature-based gray uneven image fast segmentation method. Similarity theory fast estimation bias field information is introduced, which simplifies a local information model. The running speed is greatly improved. The sensitivity to initialization contour information is reduced. Compared with a classical algorithm, the method has the advantages that a Heaviside function similar to a step function is constructed; a segmentation curve is more smooth; a dual termination condition is introduced; and according to different self-adaptive end curve evolution processes of an image content, the speed of a segmentation algorithm is improved.
Description
Technical field
The invention belongs to technical field of image processing, particularly relate to the uneven image fast segmentation method of a kind of gray scale based on multiple features.
Background technology
Owing to being subject to the impact of the unequal factor of coil frequency of illumination, shooting angle and acceptance pattern picture, some images are caused to present the uneven characteristic of gray scale, particularly particularly outstanding in medical image performance, wherein the difference of image-forming principle and own characteristic cause medical image segmentation to become a great problem.Such as in medical image, object boundary causes the image produced to have weak boundary, the characteristic such as image blurring because being subject to the impact of the factor such as low signal-to-noise ratio and bias voltage field, thus causes interference to a certain degree to Iamge Segmentation.This inequality major embodiment is the system change of the partial statistics characteristic of image.According to the feature of the uneven image of gray scale, the mathematical description form of its image is as follows:
I(x)=b(x)J(x)+n(x),x∈Ω(1)
Wherein in formula (1), I (x) represents the image of gray scale inequality, and B (x) represents the bias voltage field in the uneven region of gray scale in image, and J (x) represents true picture, and N (x) represents the noise information of image.
Being segmented in computer vision system of fast accurate plays a part key, it is the basis of graphical analysis and image understanding and identification, but because the solution of Iamge Segmentation does not have uniqueness, Iamge Segmentation is made to become an ill-posed problem, be considered to a great problem in computer vision, particularly for the medical image segmentation of gray scale inequality all the time.Therefore, need the segmentation problem finding more effective method to explore, solve medical image, this kind of ill-posed problem is converted into well-posed problem and carries out approximate solution.For such problem, there is a lot of method segmentation problem solving the uneven image of gray scale in recent years, mainly comprised the algorithm based on region model of growth; Based on the partitioning algorithm of machine learning; Based on the algorithm etc. of fuzzy set algorithm and the movable contour model based on level set, wherein based on the algorithm of level set movable contour model because namely have employed the image information of bottom, combine again high-rise priori, closer to the visual theory of the mankind, thus be widely used in the uneven medical image segmentation of gray scale.The such as C-V model etc. of more typical C-V model, local binary model of fit (LBF), regional area.But require in original method that initialization area needs near object edge, and the speed of convergence of curve is slower; The present invention is directed to this problem and carry out research work.
Summary of the invention
The present invention is responsive to initialization information in order to solve original Image Segmentation Model, and segmentation speed is slow, easily occurs the problems such as boundary leakage in image weak boundary region; Propose the uneven image fast segmentation method of a kind of gray scale based on multiple features, to reduce algorithm to the susceptibility of initialization profile and the speed and the accuracy that improve segmentation.
Technical scheme of the present invention is: the uneven image fast segmentation method of a kind of gray scale based on multiple features, comprises the steps:
Step 1, inputs image to be split: I0.
Step 2, arranges initialization closed curve profile C
0, use formula (2) initialization level set function φ
0, setup times step-length: Δ t=0.1, the optimum configurations be used in controlling curve smoothness function Heaviside is: ε=1.5, length penalty term parameter: μ=λ × 255
2, λ ∈ (0,1); Formula (3) H
newx () is line smoothing degree Heaviside function;
Wherein, π is circular constant;
Step 3, sets up local message model, global information model and regularization energy model according to initialization information in step 2, is embedded in level set framework by three class models, obtains gross energy information model and carries out curve evolvement;
Described gross energy information model is expressed as follows:
E(φ)=α·E
L(φ)+β·E
G(φ)+R
p(φ)(4)
Wherein E (φ) represents total energy model; E
l(φ) local statistic information model is represented; E
g(φ) global statistics information model is represented; R
p(φ) be used for representing distance regularization energy model; α, β are used for the nonnegative constant of controls local and global information model respectively;
Whether three category information model insertions in step 3 are carried out curve evolvement to level set framework by step 4, use dual end condition to carry out judgment curves evolution and stop: if condition does not meet, then jump to step (3) and then carry out curve evolvement; If satisfy condition, then algorithm stops, and extracts zero level collection profile and obtains segmentation result.
Further, described step 3 comprises the steps:
Step 3.1: utilize partial statistics similarity feature assessment true picture model;
Wherein make W={w
1..., w
krepresent weighting parameter, w
kfor the weights of a kth basis function of correspondence; Make G (x)={ g
1(x) ..., g
k(x) } represent basis function, g
kfor a kth basis function; K represents the number of basis function, and this method carrys out estimation function value by the subspace defining an eight neighborhood, and wherein basis function uses the gaussian kernel function of different scale to estimate to represent; Thus draw local statistic information model:
J(x)=I(x)/B'(x)(6)
Wherein J (x) represents local message model estimate value, and I (x) represents original image, bias voltage field information estimated value in B ' (x) representation formula (3);
Correspondingly, the local message model energy the Representation Equation with the image of implicit function form is as follows:
Wherein E
l(d
1, d
2, φ) and represent local message energy function; Ω represents closed curve region; d
1, d
2represent the gray average inside and outside image local area respectively; φ (x) represents implicit function zero level collection profile;
Step 3.2: utilize overall compatible characteristics to improve global energy information model;
Construct the smoothness that the high Heaviside function of degree of approximation improves closed curve, the function definition of described structure is Ru shown in above-mentioned formula (3), and its derived function is:
Described H
newfunction Fitting 0-1 step function structure new function;
Draw global energy information model below:
Wherein E
g(c
1, c
2, φ) and represent global statistics information model c
1, c
2represent the gray average inside and outside image local area I (x) respectively;
Step 3.3: the quick polynomial function of matching sets up regular terms model;
Regularization function and derived function thereof are described below:
Wherein
represent Regularization function,
represent this argument of function, its physical meaning represents symbol distance property function;
Draw to draw a conclusion by analyzing:
(a) when
time, rate of diffusion is just, then reduce
change;
(b) when
time, rate of diffusion is negative, then increase
change;
(c) when
time, diffusivity is just, then reduce further
change until be zero;
The mathematical form being provided distance regularization energy by above-mentioned analysis is expressed as follows:
Wherein λ
1, μ
1represent and control constant parameter.
Further, described dual end condition is:
The iteration controller of an integer type is set respectively in curve evolvement process, and the iteration mark of a Boolean type; Whether the difference of closed curve being used for identifying adjacent moment arrives given critical value, described iteration controller initial value is 1, the iteration mark initial value of described Boolean type is true, if reach critical value, then continue again to judge the threshold value whether iteration controller reaches given, if reach threshold value, then revise iteration and be designated vacation, curve evolvement stops; Otherwise iteration controller is from increasing, and proceeds circulation; The number of times of this method in whole circulation is less than default iteration maximum times.
Preferably, what described dual end condition was concrete is: length Len (C (t)), the Len (C (t-1)) of known t and t-1 moment zero level collection closed curve; Closed curve length threshold LenMin; Iteration maximum times Tmax; The iteration mark bool of Boolean type, when this variable-value is true, represents that algorithm proceeds iteration; When this variable-value is false, represent algorithm termination of iterations.Wherein initial value is true; Iteration count n, initial value is 1, iteration controller k when satisfying condition, and initial value is 1, threshold value Threshold=10;
Be true when meeting iteration mark bool, and when iteration count n is less than Tmax: the length difference calculating current closed curve: cur_length=|Len (C (t))-Len (C (t-1)) |; Then counter upgrades: n=n+1; If t closed curve length difference cur_length is less than closed curve length threshold LenMin, and k=Threshold, make bool=false, curve evolvement iteration ends; Otherwise k=k+1;
Be true when not meeting iteration mark bool, and when iteration count n is less than Tmax: make k=1, proceed above-mentioned iterative process.
The invention has the beneficial effects as follows: the uneven image fast segmentation method of a kind of gray scale based on multiple features, bias voltage field information is estimated fast by introducing similarity theory, thus simplify local message model, not only be greatly improved in travelling speed, and reduce the susceptibility to initialization profile information; This method, compared with classic algorithm, by the Heaviside function that structure and step function are more approximate, the smoothness of segmentation curve shows more excellent; This method, by introducing dual end condition, according to the difference of picture material adaptive end curve evolvement process, thus improves the speed of partitioning algorithm.
Accompanying drawing explanation
Fig. 1. one-piece construction process flow diagram of the present invention;
The image of the original Heaviside function of Fig. 2-1.;
The image of the function that Fig. 2-2. the present invention proposes;
The Regularization function being used for level of control collection stability that Fig. 3-1. the present invention proposes;
The derived function figure of Regularization function of level of control collection stability that what Fig. 3-2. the present invention proposed be used for;
Fig. 4. the present invention and existing classical way synthetic images segmentation result comparison diagram;
Fig. 5. the present invention and existing classical way are to x-ray shin bone image segmentation result comparison diagram;
Fig. 6. the present invention and existing classical way are to Methods of Segmentation On Cell Images Comparative result figure under microscope.
Embodiment
Below in conjunction with drawings and Examples, the invention will be further described.
This example verifies validity of the present invention by outfield experiments.
The uneven image fast segmentation method of gray scale based on multiple features, is characterized in that, comprise the steps:
Step 1, inputs image to be split: I0.
Step 2, initialization closed curve profile and correlation parameter: initial profile C is set
0, use formula (2) initialization level set function φ
0, setup times step-length: Δ t=0.1, the optimum configurations be used in controlling curve smoothness function Heaviside is: ε=1.5, length penalty term parameter: μ=λ × 255
2, λ ∈ (0,1); Formula (3) H
newx () is line smoothing degree Heaviside function, wherein the curve of this function is as shown in Fig. 2-2, Fig. 2-1 represents original Heaviside function curve diagram, the result that in this function, the value difference of parameter ε (using epsilon to replace in Fig. 2) obtains is different, value respectively for this parameter in Fig. 2-1 and Fig. 2-2 is 0.5, four value curve plottings such as 1,1.5 and 2 compare;
Wherein, π is circular constant;
Step 3, similarity principle is utilized to build local energy information model: to set up local message model, global information model and regularization energy model according to initialization information in step 2, three class models are embedded in level set framework, obtain gross energy information model and carry out curve evolvement;
Described gross energy information model is expressed as follows:
E(φ)=α·E
L(φ)+β·E
G(φ)+R
p(φ)(4)
Wherein E (φ) represents total energy model; E
l(φ) local statistic information model is represented; E
g(φ) global statistics information model is represented; R
p(φ) be used for representing distance regularization energy model; α, β are used for the nonnegative constant of controls local and global information model respectively;
Step 3.1: utilize partial statistics similarity feature assessment true picture model;
Wherein make W={w
1..., w
krepresent weighting parameter, w
kfor the weights of a kth basis function of correspondence; Make G (x)={ g
1(x) ..., g
k(x) } represent basis function, g
kfor a kth basis function; K represents the number of basis function, and this method carrys out estimation function value by the subspace defining an eight neighborhood, and wherein basis function uses the gaussian kernel function of different scale to estimate to represent; Thus draw local statistic information model:
J(x)=I(x)/B'(x)(6)
Wherein J (x) represents local message model estimate value, and I (x) represents original image, bias voltage field information estimated value in B ' (x) representation formula (3);
Correspondingly, the local message model energy the Representation Equation with the image of implicit function form is as follows:
Wherein E
l(d
1, d
2, φ) and represent local message energy function; Ω represents closed curve region; d
1, d
2represent the gray average inside and outside image local area respectively; φ (x) represents implicit function zero level collection profile;
Step 3.2: utilize overall compatible characteristics to improve global energy information model;
Construct the smoothness that the high Heaviside function of degree of approximation improves closed curve, the function definition of described structure is Ru shown in above-mentioned formula (3), and its derived function is:
Described H
newfunction Fitting 0-1 step function structure new function;
Draw global energy information model below:
Wherein E
g(c
1, c
2, φ) and represent global statistics information model c
1, c
2represent the gray average inside and outside image local area I (x) respectively;
Step 3.3: the quick polynomial function of matching sets up regular terms model;
Regularization function and derived function thereof are described below:
Wherein
represent Regularization function, as shown in figure 3-1,
represent this argument of function, its physical meaning represents symbol distance property function;
function as shown in figure 3-2,
represent this argument of function.
Analyzed by Fig. 3-1 and Fig. 3-2 and draw to draw a conclusion:
(a) when
time, rate of diffusion is just, then reduce
change;
(b) when
time, rate of diffusion is negative, then increase
change;
(c) when
time, diffusivity is just, then reduce further
change until be zero;
The mathematical form being provided distance regularization energy by above-mentioned analysis is expressed as follows:
Wherein λ
1, μ
1represent and control constant parameter.
Whether three category information model insertions in step 3 are carried out curve evolvement to level set framework by step 4, use dual end condition to carry out judgment curves evolution and stop: if condition does not meet, then jump to step (3) and then carry out curve evolvement; If satisfy condition, then algorithm stops, and extracts zero level collection profile and obtains segmentation result.
Described dual end condition is:
The iteration controller of an integer type is set respectively in curve evolvement process, and the iteration mark of a Boolean type; Whether the difference of closed curve being used for identifying adjacent moment arrives given critical value, described iteration controller initial value is 1, the iteration mark initial value of described Boolean type is true, if reach critical value, then continue again to judge the threshold value whether iteration controller reaches given, if reach threshold value, then revise iteration and be designated vacation, curve evolvement stops; Otherwise iteration controller is from increasing, and proceeds circulation; The number of times of this method in whole circulation is less than default iteration maximum times.
What described dual end condition was concrete is: length Len (C (t)), the Len (C (t-1)) of known t and t-1 moment zero level collection closed curve; Closed curve length threshold LenMin; Iteration maximum times Tmax; The iteration mark bool of Boolean type, when this variable-value is true, represents that algorithm proceeds iteration; When this variable-value is false, represent algorithm termination of iterations.Wherein initial value is true; Iteration count n, initial value is 1, iteration controller k when satisfying condition, and initial value is 1, threshold value Threshold=10;
Be true when meeting iteration mark bool, and when iteration count n is less than Tmax: the length difference calculating current closed curve: cur_length=|Len (C (t))-Len (C (t-1)) |; Then counter upgrades: n=n+1; If t closed curve length difference cur_length is less than closed curve length threshold LenMin, and k=Threshold, make bool=false, curve evolvement iteration ends; Otherwise k=k+1;
Be true when not meeting iteration mark bool, and when iteration count n is less than Tmax: make k=1, proceed above-mentioned iterative process.
Effect of the present invention can be verified by following contrast experiment:
1. description of test and experiment condition are arranged
(1) description of test
In Fig. 4-Fig. 6, (a)-(e) represents respectively: (a) two kinds of initialization profiles: the first row image represents initialization single window image; Second row image represents initialization two video in windows; (b) segmentation result of the present invention; C C-V model segmentation result that () calendar year 2001 is proposed by Chan and Vase; D nonlinear adaptive Level Set Method (NLAL) segmentation result that () proposes for 2014; (e) 2008 RSF (region-scalablefittingenergy) method segmentation result of classics of proposing of the people such as Nian Li.
Fig. 4: with the composograph of Gaussian noise;
Fig. 5: X ray skeletal graph picture: left side is shin bone, right side is fibula;
Fig. 6: cell image under microscope
(2) experiment condition is arranged
According to the value of epsilon in analysis chart 2, the parameter ε in whole experiment Heaviside function is all set to 1.5 can ensure that the smoothness splitting curve is more excellent; Optimum configurations wherein in formula (4) (12) is as follows:
Figure (4): α=0.3, β=1.0, λ=1, μ=0.005 × 255
2
Figure (5): α=0.6, β=1.0, λ=1, μ=0.012 × 255
2
Figure (6): α=0.6, β=0.8, λ=1, μ=0.002 × 255
2
2. experiment content and result:
For Fig. 4: when using C-V model method segmentation composograph profile variations comparatively large regions there is oscillatory occurences; Use NLAL model method correctly cannot be partitioned into the larger image of the profile variations of original image; Use RSF model segmentation result to show, carrying out easily being absorbed in local optimum in iterative process, thus causing segmentation effect undesirable, the method is more responsive to initialized location, and the inventive method segmentation result is more accurate.
For Fig. 5: use C-V model method and the inventive method to carry out splitting and all obtain satisfied effect, (chronomere calculated with second) as shown in table 1: the inventive method is far superior to the C-V model method (the method is primarily of two people's namings) that calendar year 2001 proposes by people such as Chan and Vese in speed.NLAL model method has only been partitioned into shin bone part preferably; And typical RSF model method shows Expired Drugs.
Can partitioning portion region for Fig. 6: NLAL method, trend towards for oval image for segmentation contour, the method is more effective; The RSF method that the people such as contrast Lee propose is more responsive to initialization; Classical C-V model method has Expired Drugs; The inventive method is other three Modeling Approaches comparatively, and its segmentation result is good.
Table 1 each method convergence time comparing result
Claims (4)
1., based on the uneven image fast segmentation method of gray scale of multiple features, it is characterized in that, comprise the steps:
Step 1, inputs image to be split: I0;
Step 2, arranges initial profile C
0, use formula (2) initialization level set function φ
0, setup times step-length: Δ t=0.1, the optimum configurations be used in controlling curve smoothness function Heaviside is: ε=1.5, length penalty term parameter: μ=λ × 255
2, λ ∈ (0,1); Formula (3) H
newx () is line smoothing degree Heaviside function;
Wherein, π is circular constant;
Step 3, sets up local message model, global information model and regularization energy model according to initialization information in step 2, is embedded in level set framework by three class models, obtains gross energy information model and carries out curve evolvement;
Described gross energy information model is expressed as follows:
E(φ)=α·E
L(φ)+β·E
G(φ)+R
p(φ)(4)
Wherein E (φ) represents total energy model; E
l(φ) local statistic information model is represented; E
g(φ) global statistics information model is represented; R
p(φ) be used for representing distance regularization energy model; α, β are used for the nonnegative constant of controls local and global information model respectively;
Whether three category information model insertions in step 3 are carried out curve evolvement to level set framework by step 4, use dual end condition to carry out judgment curves evolution and stop: if condition does not meet, then jump to step (3) and then carry out curve evolvement; If satisfy condition, then algorithm stops, and extracts zero level collection profile and obtains segmentation result.
2. the uneven image fast segmentation method of a kind of gray scale based on multiple features according to claim 1, it is characterized in that, described step 3 comprises the steps:
Step 3.1: utilize partial statistics similarity feature assessment true picture model;
Wherein make W={w
1..., w
krepresent weighting parameter, w
kfor the weights of a kth basis function of correspondence; Make G (x)={ g
1(x) ..., g
k(x) } represent basis function, g
kfor a kth basis function; K represents the number of basis function, and this method carrys out estimation function value by the subspace defining an eight neighborhood, and wherein basis function uses the gaussian kernel function of different scale to estimate to represent; Thus draw local statistic information model:
J(x)=I(x)/B'(x)(6)
Wherein J (x) represents local message model estimate value, and I (x) represents original image, bias voltage field information estimated value in B ' (x) representation formula (3);
Correspondingly, the local message model energy the Representation Equation with the image of implicit function form is as follows:
E
L(d
1,d
2,φ)=∫
Ω|I(x)/B'(x)-d
1|
2H
new(φ(x))dx+
(7)
∫
Ω|I(x)/B'(x)-d
2|
2(1-H
new(φ(x)))dx
Wherein E
l(d
1, d
2, φ) and represent local message energy function; Ω represents closed curve region; d
1, d
2represent the gray average inside and outside image local area respectively; φ (x) represents implicit function zero level collection profile;
Step 3.2: utilize overall compatible characteristics to improve global energy information model;
Construct the smoothness that the high Heaviside function of degree of approximation improves closed curve, the function definition of described structure is Ru shown in above-mentioned formula (3), and its derived function is:
Described H
newfunction Fitting 0-1 step function structure new function;
Draw global energy information model below:
E
G(c
1,c
2,φ)=∫
Ω|I(x)-c
1|
2H
new(φ(x))dx+
(9)
∫
Ω|I(x)-c
2|
2(1-H
new(φ(x)))dx
Wherein E
g(c
1, c
2, φ) and represent global statistics information model c
1, c
2represent the gray average inside and outside image local area I (x) respectively;
Step 3.3: the quick polynomial function of matching sets up regular terms model;
Regularization function and derived function thereof are described below:
Wherein p (| ▽ φ |) represents Regularization function, | ▽ φ | represent this argument of function, its physical meaning represents symbol distance property function;
Draw to draw a conclusion by analyzing:
A () is as | ▽ φ | during > 1, rate of diffusion is just, then reduce | ▽ φ | change;
(b) when
time, rate of diffusion is negative, then increase | ▽ φ | change;
(c) when
time, diffusivity is just, then further reduce | ▽ φ | change until be zero;
The mathematical form being provided distance regularization energy by above-mentioned analysis is expressed as follows:
E
R=λ
1·∫
Ωδ(φ(x))dx+μ
1·∫
Ωp(|▽φ(x)|)dx(12)
Wherein λ
1, μ
1represent and control constant parameter.
3. the uneven image fast segmentation method of a kind of gray scale based on multiple features according to claim 1, it is characterized in that, described dual end condition is:
The iteration controller of an integer type is set respectively in curve evolvement process, and the iteration mark of a Boolean type; Whether the difference of closed curve being used for identifying adjacent moment arrives given critical value, described iteration controller initial value is 1, the iteration mark initial value of described Boolean type is true, if reach critical value, then continue again to judge the threshold value whether iteration controller reaches given, if reach threshold value, then revise iteration and be designated vacation, curve evolvement stops; Otherwise iteration controller is from increasing and proceeding circulation; The number of times of this method in whole circulation is less than default iteration maximum times.
4. the uneven image fast segmentation method of a kind of gray scale based on multiple features according to claim 1, it is characterized in that, what described dual end condition was concrete is: length Len (C (t)), the Len (C (t-1)) of known t and t-1 moment zero level collection closed curve; Closed curve length threshold LenMin; Iteration maximum times Tmax; The iteration mark bool of Boolean type, when this variable-value is true, represents that algorithm proceeds iteration; When this variable-value is false, represent algorithm termination of iterations; Wherein initial value is true; Iteration count n, initial value is 1, iteration controller k when satisfying condition, and initial value is 1, threshold value Threshold=10;
Be true when meeting iteration mark bool, and when iteration count n is less than Tmax: the length difference calculating current closed curve: cur_length=|Len (C (t))-Len (C (t-1)) |; Then counter upgrades: n=n+1; If t closed curve length difference cur_length is less than closed curve length threshold LenMin, and k=Threshold, make bool=false, curve evolvement iteration ends; Otherwise k=k+1;
Be true when not meeting iteration mark bool, and when iteration count n is less than Tmax: make k=1, proceed above-mentioned iterative process.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510531170.7A CN105139398A (en) | 2015-08-26 | 2015-08-26 | Multi-feature-based gray uneven image fast segmentation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510531170.7A CN105139398A (en) | 2015-08-26 | 2015-08-26 | Multi-feature-based gray uneven image fast segmentation method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105139398A true CN105139398A (en) | 2015-12-09 |
Family
ID=54724730
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510531170.7A Pending CN105139398A (en) | 2015-08-26 | 2015-08-26 | Multi-feature-based gray uneven image fast segmentation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105139398A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105844641A (en) * | 2016-03-24 | 2016-08-10 | 武汉工程大学 | Adaptive threshold segmentation method in dynamic environment |
CN106548478A (en) * | 2016-10-28 | 2017-03-29 | 中国科学院苏州生物医学工程技术研究所 | Active contour image partition method based on local fit image |
CN107798684A (en) * | 2017-11-07 | 2018-03-13 | 河南师范大学 | A kind of active contour image partition method and device based on symbol pressure function |
CN109785293A (en) * | 2018-12-22 | 2019-05-21 | 昆明理工大学 | A method of the ultrasound image lesion based on global and local movable contour model is divided automatically |
CN111091578A (en) * | 2019-12-17 | 2020-05-01 | 湖北民族大学 | Segmentation method for blood vessel medical image |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2002056241A1 (en) * | 2001-01-12 | 2002-07-18 | University Of Florida | Computational algorithms for image registration |
CN102354396A (en) * | 2011-09-23 | 2012-02-15 | 清华大学深圳研究生院 | Method for segmenting image with non-uniform gray scale based on level set function |
-
2015
- 2015-08-26 CN CN201510531170.7A patent/CN105139398A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2002056241A1 (en) * | 2001-01-12 | 2002-07-18 | University Of Florida | Computational algorithms for image registration |
CN102354396A (en) * | 2011-09-23 | 2012-02-15 | 清华大学深圳研究生院 | Method for segmenting image with non-uniform gray scale based on level set function |
Non-Patent Citations (3)
Title |
---|
XIAO-FENG WANG 等: "An efficient local Chan–Vese model for image segmentation", 《PATTERN RECOGNITION》 * |
XIAO-FENGWANG 等: "A novellevelsetmethodforimagesegmentationbyincorporating local statisticalanalysisandglobalsimilaritymeasurement", 《PATTERN RECOGNITION》 * |
YU HAIPING 等: "Regularized Level Set Method by Incorporating Local Statistical Information and Global Similarity Compatibility for Image Segmentation", 《INTERNATIONAL CONFERENCE ON INTELLIGENT COMPUTING》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105844641A (en) * | 2016-03-24 | 2016-08-10 | 武汉工程大学 | Adaptive threshold segmentation method in dynamic environment |
CN105844641B (en) * | 2016-03-24 | 2019-05-28 | 武汉工程大学 | A kind of adaptive threshold fuzziness method under dynamic environment |
CN106548478A (en) * | 2016-10-28 | 2017-03-29 | 中国科学院苏州生物医学工程技术研究所 | Active contour image partition method based on local fit image |
CN106548478B (en) * | 2016-10-28 | 2019-06-28 | 中国科学院苏州生物医学工程技术研究所 | Active contour image partition method based on local fit image |
CN107798684A (en) * | 2017-11-07 | 2018-03-13 | 河南师范大学 | A kind of active contour image partition method and device based on symbol pressure function |
CN107798684B (en) * | 2017-11-07 | 2020-02-28 | 河南师范大学 | Active contour image segmentation method and device based on symbol pressure function |
CN109785293A (en) * | 2018-12-22 | 2019-05-21 | 昆明理工大学 | A method of the ultrasound image lesion based on global and local movable contour model is divided automatically |
CN109785293B (en) * | 2018-12-22 | 2022-09-27 | 昆明理工大学 | Ultrasonic image focus segmentation method based on global and local active contour models |
CN111091578A (en) * | 2019-12-17 | 2020-05-01 | 湖北民族大学 | Segmentation method for blood vessel medical image |
CN111091578B (en) * | 2019-12-17 | 2023-07-04 | 湖北民族大学 | Segmentation method for vascular medical image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105139398A (en) | Multi-feature-based gray uneven image fast segmentation method | |
US20220414891A1 (en) | Method for automatic segmentation of fuzzy boundary image based on active contour and deep learning | |
Wang | Stochastic relaxation on partitions with connected components and its application to image segmentation | |
CN101447076B (en) | Method for partitioning interested areas in WEB image | |
CN103761726B (en) | Block adaptive image partition method based on FCM | |
CN103942794A (en) | Image collaborative cutout method based on confidence level | |
CN103150731B (en) | A kind of fuzzy clustering image partition method | |
CN111047603B (en) | Aerial image hybrid segmentation algorithm based on novel Markov random field and region combination | |
CN108509866A (en) | A kind of facial contour extraction method | |
CN108460781B (en) | Active contour image segmentation method and device based on improved SPF | |
CN104835168B (en) | Quick multiphase image dividing method based on global convex optimization Variation Model | |
Rudolph et al. | Structuring autoencoders | |
CN110728302A (en) | Method for identifying color textile fabric tissue based on HSV (hue, saturation, value) and Lab (Lab) color spaces | |
CN107123130A (en) | Kernel correlation filtering target tracking method based on superpixel and hybrid hash | |
CN105469408A (en) | Building group segmentation method for SAR image | |
CN104732551A (en) | Level set image segmentation method based on superpixel and graph-cup optimizing | |
Yuan et al. | Watershed-based superpixels with global and local boundary marching | |
CN109242876A (en) | A kind of image segmentation algorithm based on markov random file | |
CN106127112A (en) | Data Dimensionality Reduction based on DLLE model and feature understanding method | |
CN107424153B (en) | Face segmentation method based on deep learning and level set | |
CN101430789B (en) | Image edge detection method based on Fast Slant Stack transformation | |
CN110473224B (en) | Automatic RSF level set image segmentation method based on KL entropy | |
CN107798684B (en) | Active contour image segmentation method and device based on symbol pressure function | |
CN108665470B (en) | Interactive contour extraction method | |
CN114596320A (en) | Image segmentation method and device based on ALSLCV model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into 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: 20151209 |