CN112102243A - Active contour segmentation method and system combining general energy function and prior information item - Google Patents

Active contour segmentation method and system combining general energy function and prior information item Download PDF

Info

Publication number
CN112102243A
CN112102243A CN202010811710.8A CN202010811710A CN112102243A CN 112102243 A CN112102243 A CN 112102243A CN 202010811710 A CN202010811710 A CN 202010811710A CN 112102243 A CN112102243 A CN 112102243A
Authority
CN
China
Prior art keywords
energy function
image
function
segmentation
prior information
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
CN202010811710.8A
Other languages
Chinese (zh)
Other versions
CN112102243B (en
Inventor
杨云云
王若凡
冯翀
谢睿诚
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.)
Shenzhen Graduate School Harbin Institute of Technology
Original Assignee
Shenzhen Graduate School Harbin Institute of Technology
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 Shenzhen Graduate School Harbin Institute of Technology filed Critical Shenzhen Graduate School Harbin Institute of Technology
Priority to CN202010811710.8A priority Critical patent/CN112102243B/en
Publication of CN112102243A publication Critical patent/CN112102243A/en
Application granted granted Critical
Publication of CN112102243B publication Critical patent/CN112102243B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention provides an active contour segmentation method and system combining a general energy function and a prior information item, wherein the active contour segmentation method comprises the following steps: step 1: selecting an acquisition mode of prior information to obtain a segmentation result of the fuzzy clustering; step 2: inputting an image to be segmented, an initial level set function, a stopping condition, an initial Bregman variable and an initial auxiliary variable; and step 3: calculating the image approximate intensity inside and outside the initial contour line; and 4, step 4: calculating an image data fitting term; and 5: calculating a level set function; step 6: calculating an auxiliary vector and a Bregman variable; and 7: and (6) iteration. The invention has the beneficial effects that: 1. the invention uses the prior information, so that the accuracy of the result is greatly improved, and the automatically acquired prior information can effectively reduce manual operation and improve the working efficiency.

Description

Active contour segmentation method and system combining general energy function and prior information item
Technical Field
The invention relates to the field of image segmentation and processing, in particular to an active contour segmentation method and system combining a general energy function and a prior information item.
Background
The image segmentation has a wide application foundation in the fields of industrial production and medical civilian life, can effectively improve the level of industrial automation, and realizes the positioning and navigation of targets in the traffic field. In the field of medical imaging, doctors are assisted in performing operations and positioning lesion areas.
In the field of medical image processing, an image segmentation algorithm effectively segments ultrasonic images and nuclear magnetic resonance images, can distinguish different organs and tissues, marks names, assists doctors in diagnosing related diseases, improves the diagnosis efficiency of medical departments, and popularizes experience and practice to a wider basic medical system. In the traffic field, the image segmentation technology is applied to the automatic driving field, so that road obstacles are identified, and the driving safety and efficiency are improved.
At present, the existing image segmentation methods can be divided into two types, namely a boundary detection method, a region growth method, an active contour model and other traditional methods and a latest machine learning method.
The boundary detection method searches the position where the pixel value at the edge is discontinuous through a boundary detection operator, and the position is called as a sudden change position. Such discontinuities can be detected by taking the derivative, for step-like edges, whose position corresponds to the extreme point of the first derivative and to the zero point of the second derivative. Typical operators include LoG operators, Canny operators, and the like.
The region growing method combines adjacent pixels or regions with similar growing point properties with the growing points according to the pixel values to form new growing points, and repeats the process until the growing points cannot grow.
The RSF model commonly used in active contour models is also a region segmentation based model. The method sets omega as the image definition domain,
Figure BDA0002631206870000011
as a gray scale image, f1,f2Are two functions representing the image intensity inside and outside the segmentation boundary. The task of the model is to minimize the energy function, so as to find an approximate image u and the approximate image uBoundary C, thereby dividing the image into two parts.
The latest deep learning method has been widely applied to the field of image segmentation since the advent. Different networks may handle different segmentation problems. Earlier image segmentation methods have more to find edges (lines and curves) or gradient elements. Now, the concept of semantic segmentation will allow computers to provide pixel-level image understanding in a fully human-perceptible manner. Semantic segmentation solves this problem by clustering together image portions belonging to the same target, thereby expanding its application field.
R-CNN is the first algorithm to apply a deep neural network to target detection, but the efficiency is too low, and many improvements are made by later people. For example, full consideration of detail information, rebalancing of weights in median use for prediction of classes, and so forth.
FCN is an upsampling based method, where in a deconvolution, upsampling structure, a picture is first upsampled (to enlarge pixels) and then convolved. This reduces the loss of image detail information during convolution. The U-NET is a network structure designed for medical image segmentation. Aiming at the particularity of medical images, the method provided by the U-NET effectively improves the effect of training and detecting by using a small amount of data sets, and simultaneously provides an effective method for processing large-size images, so that the method is a network structure which is widely used at present.
The boundary detection algorithm has a simple mathematical model, is popular and easy to understand, but has a poor segmentation result, only finds part of the contour boundary, is easily interfered by other areas in the image, and generates a segmentation result which exists outside a target area of a plurality of objects and in the background. The limitations of the region growing method are the selection of the initial growing point pixels and the establishment of the growing criteria. These two points will have a great influence on the segmentation result.
The RSF model is sensitive to initialization, especially the initial contour and various parameters in the model have a large effect on the result. Therefore, a good result is obtained by utilizing the RSF model, the initial contour needs to be selected according to different image characteristics, parameters are adjusted through multiple tests, the labor and the time are wasted, and a large promotion space is provided.
The deep learning method still has a plurality of defects as the image segmentation algorithm which is widely used at present. Firstly, mass data is needed for training a model, and a medical image is different from a natural image and is extremely difficult to obtain, and a label corresponding to the image, namely a lesion position, needs to be marked by a professional doctor, so that the scarcity of medical image resources is further reflected. Second, the deep learning algorithm is highly dependent on computer hardware facilities, which brings a huge cost investment. In addition, the specificity of medical images puts higher demands on the configuration of the computer. This does not serve to accurately diagnose diseases if computational efficiency is pursued by reducing image quality.
Disclosure of Invention
The invention provides an active contour segmentation method combining a general energy function and a prior information item, which comprises the following steps:
step 1: selecting an acquisition mode of prior information to obtain a segmentation result of the fuzzy clustering as input psispi(ii) a Selecting an energy function as Hgen(Ψ)；
Step 2: inputting an image I to be segmented, and an initial level set function psi0Stop condition gamma, initial Bregman variable b00 and initial auxiliary variable h0＝0；
And step 3: calculating the approximate intensity of the image inside and outside the initial contour line
Figure BDA0002631206870000021
And 4, step 4: calculating an image data fitting term Sk；
And 5: computing the level set function Ψk+1；
Step 6: calculating an auxiliary vector hk+1And Bregman variable bk+1；
And 7: iteration
Figure BDA0002631206870000031
And 8: judge | | | Ψk+1-ΨkLess than or equal to gamma; if | | | Ψk+1-ΨkIf | | ≦ γ, executing step 9, otherwise returning to step 4, and continuing to calculate image data fitting item Sk；
And step 9: outputting the final result, i.e. the segmentation profile Zk+1。
As a further improvement of the present invention, in the step 1, the following steps are further performed:
step S1: establishing a general image segmentation energy function based on prior information;
step S2, defining a prior information item and selecting an energy function item;
and step S3, selecting specific prior information to minimize the energy functional.
As a further improvement of the present invention, in step S1, the method specifically includes:
consider an image
Figure BDA0002631206870000032
Is the level set function over the domain D, we define an energy function H (ψ):
Figure BDA0002631206870000033
wherein Hgen(Ψ) refers to a general energy function, and we select a boundary-based energy function, a region-based energy function or other type of energy function, H, based on the characteristics and needs of the different imagesspi(psi) refers to a general a priori information item.
As a further improvement of the present invention, in step S2, the method specifically includes:
defining a priori information item and selecting an energy function item, the prior information item Hspi(Ψ) we define as:
Hspi(Ψ)＝∫D|Ψ(x)-Ψspi(x)|2dx, (2)
therein ΨspiLevel set function representing the result of the prior information segmentation, Ψ representing the evolvingLevel set function, for the energy function term, we choose to take a region-based energy function as an example, which is expressed in terms of inner product:
Figure BDA0002631206870000034
s is the image data fit term, ω is a boundary detection operator defined as:
Figure BDA0002631206870000035
in the formula, x and y represent the positions of pixel points in the image, and lambda1,λ2Beta are all parameters greater than zero, GσIs a Gaussian kernel function, σ is the standard deviation, v1,v2Respectively, the approximate image intensities of the inside and outside of the active contour.
As a further improvement of the present invention, in step S3, the method specifically includes:
according to the formulas (2) and (3), the minimization problem of the model is as follows:
Figure BDA0002631206870000041
to solve this minimization problem, first, the Bregman variable b ═ b (b) is introducedx,by) And an auxiliary variable h ═ h (h)x,hy) So that
Figure BDA0002631206870000042
Then minimization problem equation (5) is solved by the following equation:
Figure BDA0002631206870000043
wherein psik+1Is a level set function, hk+1Is an auxiliary variable, bk+1And bkIs the Bregman variable;
finally, we denote the evolving activity profile by z, defined as:
zk+1＝{x:Ψk+1(x)＝0}. (7)
the invention also discloses an active contour segmentation system combining the general energy function and the prior information item, which comprises the following steps:
a selection unit: selecting the acquisition mode of prior information to obtain the segmentation result of fuzzy clustering as input psispi(ii) a Selecting an energy function as Hgen(Ψ)；
An input unit: for inputting an image I to be segmented, an initial level set function Ψ0Stop condition gamma, initial Bregman variable b00 and initial auxiliary variable h0＝0；
The first calculation unit: for calculating approximate intensities of images inside and outside initial contour lines
Figure BDA0002631206870000044
A second calculation unit: for calculating the fitting term S of image datak；
A third calculation unit: for computing the level set function Ψk+1；
A fourth calculation unit: for calculating auxiliary vectors hk+1And Bregman variable bk+1；
An iteration unit: for iteration
Figure BDA0002631206870000045
A judging unit: for judging | | Ψk+1-ΨkLess than or equal to gamma; if | | | Ψk+1-ΨkIf | | ≦ γ, executing step 9, otherwise returning to step 4, and continuing to calculate image data fitting item Sk；
An output unit: for outputting the final result-segmenting contour zk+1。
As a further improvement of the present invention, in the selection unit, the method further includes:
a building module: the image segmentation energy function is used for establishing a general image segmentation energy function based on prior information;
a definition module: for defining a priori information term and selecting an energy function term;
a minimization module: the method is used for selecting specific prior information and minimizing the energy functional.
As a further improvement of the present invention, the establishing module specifically includes:
consider an image
Figure BDA0002631206870000051
Is the level set function over the domain D, we define an energy function H (ψ):
Figure BDA0002631206870000052
wherein Hgen(Ψ) refers to a general energy function, and we select a boundary-based energy function, a region-based energy function or other type of energy function, H, based on the characteristics and needs of the different imagesspi(psi) refers to a general a priori information item.
As a further improvement of the present invention, in the definition module, the definition module specifically includes:
defining a priori information item and selecting an energy function item, the prior information item Hspi(Ψ) we define as:
Hspi(Ψ)＝∫D|Ψ(x)-Ψspi(x)|2dx, (2)
therein ΨspiLevel set function representing the result of the prior information segmentation, Ψ represents the level set function being evolved, and for the energy function term, we choose to take a region-based energy function as an example, and the energy function is expressed in the form of an inner product:
Figure BDA0002631206870000053
s is the image data fit term, ω is a boundary detection operator defined as:
Figure BDA0002631206870000054
in the formula, x and y represent the positions of pixel points in the image, and lambda1,λ2Beta are all parameters greater than zero, GσIs a Gaussian kernel function, σ is the standard deviation, v1,v2Respectively, the approximate image intensities of the inside and outside of the active contour.
As a further improvement of the present invention, the miniaturization module specifically includes:
according to the formulas (2) and (3), the minimization problem of the model is as follows:
Figure BDA0002631206870000055
to solve this minimization problem, first, the Bregman variable b ═ b (b) is introducedx,by) And an auxiliary variable h ═ h (h)x,hy) So that
Figure BDA0002631206870000056
Then minimization problem equation (5) is solved by the following equation:
Figure BDA0002631206870000061
wherein psik+1Is a level set function, hk+1Is an auxiliary variable, bk+1And bkIs the Bregman variable;
finally, we denote the evolving activity profile by z, defined as:
zk+1＝{x:Ψk+1(x)＝0}. (7)
the invention has the beneficial effects that: 1. in addition, the prior information automatically acquired by the active contour segmentation method can effectively reduce manual operation and improve the working efficiency; 2. because the general energy function is obtained from various existing mature algorithms, the active contour segmentation method of the invention not only increases the applicability to different types of images, but also increases the output of segmentation results; 3. the active contour segmentation method of the invention subsequently uses a split Bregman method to solve the minimized energy functional, compared with the existing gradient descent method, the algorithm is robust to initialization; 4. the active contour segmentation method uses the convex function as the energy function, so that the local optimal solution can not be trapped in the process of minimizing the energy function, the segmentation result is accurate, and meanwhile, under the finite iteration, the speed of minimizing the energy functional by using the split Bregman method is higher than that of the traditional gradient descent method, so that the efficiency of the active contour segmentation method is greatly improved.
Drawings
FIG. 1 is a flow chart of an active contour segmentation method of the present invention;
fig. 2 is a flow chart of the present invention for selecting prior information.
Detailed Description
As shown in FIG. 1, the invention discloses an active contour segmentation method combining a general energy function and a prior information item, which comprises the following steps:
step 1: selecting an acquisition mode of prior information, taking fuzzy clustering as the prior information as an example, acquiring a segmentation result of the fuzzy clustering as input psispi(ii) a Selecting an energy function as Hgen(Ψ)；
Step 2: inputting an image I to be segmented, and an initial level set function psi0Stop condition gamma, initial Bregman variable b00 and initial auxiliary variable h0＝0；
And step 3: calculating the approximate intensity of the image inside and outside the initial contour line
Figure BDA0002631206870000062
And 4, step 4: computing image data fit termsSk；
And 5: computing the level set function Ψk+1；
Step 6: calculating an auxiliary vector hk+1And Bregman variable bk+1；
And 7: iteration
Figure BDA0002631206870000063
And 8: judge | | | Ψk+1-ΨkLess than or equal to gamma; if | | | Ψk+1-ΨkIf | | ≦ γ, executing step 9, otherwise returning to step 4, and continuing to calculate image data fitting item SkUntil the judgment formula of the step 8 is established;
and step 9: outputting the final result segmentation contour Zk+1。
The invention discloses a method for segmenting an active contour, which is a rapid algorithm for providing a series of flexible medical image disease diagnosis and treatment and natural image target identification segmentation, and aims to solve the difficult problem of various image segmentation and effectively improve the segmentation accuracy, as shown in figure 2, in the step 1, the following steps are also executed:
step S1: establishing a general image segmentation energy function based on prior information;
in step S1, the method specifically includes:
consider an image
Figure BDA0002631206870000073
Is the level set function over the domain D, we define an energy function H (ψ):
Figure BDA0002631206870000071
wherein Hgen(Ψ) refers to a general energy function, and we can choose a boundary-based energy function, a region-based energy function or other type of energy function, H, according to the characteristics and needs of different imagesspi(psi) means oneThe general prior information item is related to the shape of the target and subjective experience of people, so that the prior information item can be obtained automatically or manually. Step S2, defining prior information item and selecting energy function item, taking an energy function based on region as an example;
in step S2, the method specifically includes:
defining a priori information item and selecting an energy function item, taking a region-based energy function as an example, the prior information item Hspi(Ψ) we define as:
Hspi(Ψ)＝∫D|Ψ(x)-Ψspi(x)|2dx, (2)
therein ΨspiLevel set function representing the result of the prior information segmentation, Ψ represents the level set function being evolved, and for the energy function term, we choose to take a region-based energy function as an example, and the energy function is expressed in the form of an inner product:
Figure BDA0002631206870000072
s is the image data fit term, ω is a boundary detection operator defined as:
Figure BDA0002631206870000081
in the formula, x and y represent the positions of pixel points in the image, and lambda1,λ2Beta are all parameters greater than zero, GσIs a Gaussian kernel function, σ is the standard deviation, v1,v2Respectively, the approximate image intensities of the inside and outside of the active contour. And step S3, selecting specific prior information, taking fuzzy clustering as the prior information as an example, and minimizing an energy functional.
In step S3, the method specifically includes:
according to the formulas (2) and (3), the minimization problem of the model is as follows:
Figure BDA0002631206870000082
to solve this minimization problem, first, the Bregman variable b ═ b (b) is introducedx,by) And an auxiliary variable h ═ h (h)x,hy) So that
Figure BDA0002631206870000083
Then minimization problem equation (5) can be solved by the following equation:
Figure BDA0002631206870000084
wherein psik+1Is a level set function, hk+1Is an auxiliary variable, bk+1And bkIs the Bregman variable;
finally, we denote the evolving activity profile by z, defined as:
zk+1＝{x:Ψk+1(x)＝0}. (7)
the invention relates to a segmentation result and model evaluation, in particular to an active contour segmentation method combining a general energy function and a prior information item, which can segment partial medical images including melanoma images, heart images, human brain images and the like and can be used for auxiliary diagnosis of related diseases. The active contour segmentation method is mainly used for comparing and analyzing segmentation results of a traditional segmentation model and a deep learning model. On the basis of quantitative evaluation indexes, an F measure is selected as an index for measuring the quality of a segmentation result. Taking the melanoma image dataset as an example, the F values of our method were increased by 81.20%, 18.93%, 8.16% and 5.02% compared to the RSF, CV, FCM and U-Net methods, respectively. In time, the active contour segmentation method provided by the invention is time-consuming and stable and is not disturbed by parameters.
The invention also discloses an active contour segmentation system combining the general energy function and the prior information item, which comprises the following steps:
a selection unit: obtaining mode for selecting prior information to obtain fuzzy clustering markCutting the result as input Ψspi(ii) a Selecting an energy function as Hgen(Ψ)；
An input unit: for inputting an image I to be segmented, an initial level set function Ψ0Stop condition gamma, initial Bregman variable b00 and initial auxiliary variable h0＝0；
The first calculation unit: for calculating approximate intensities of images inside and outside initial contour lines
Figure BDA0002631206870000091
A second calculation unit: for calculating the fitting term S of image datak；
A third calculation unit: for computing the level set function Ψk+1；
A fourth calculation unit: for calculating auxiliary vectors hk+1And Bregman variable bk+1；
An iteration unit: for iteration
Figure BDA0002631206870000092
A judging unit: for judging | | Ψk+1-ΨkLess than or equal to gamma; if | | | Ψk+1-ΨkIf | | ≦ γ, executing step 9, otherwise returning to step 4, and continuing to calculate image data fitting item Sk；
An output unit: for outputting the final result-segmenting contour Zk+1zk+1。
In the selection unit, further comprising:
a building module: the image segmentation energy function is used for establishing a general image segmentation energy function based on prior information;
a definition module: for defining a priori information term and selecting an energy function term;
a minimization module: the method is used for selecting specific prior information and minimizing the energy functional.
In the establishing module, the method specifically includes:
consider an image
Figure BDA0002631206870000093
Is the level set function over the domain D, we define an energy function H (ψ):
Figure BDA0002631206870000094
wherein Hgen(Ψ) refers to a general energy function, and we can choose a boundary-based energy function, a region-based energy function or other type of energy function, H, according to the characteristics and needs of different imagesspi(psi) refers to a general item of a priori information, both related to the morphology of the object itself and to the subjective experience of the person, so that it can be obtained automatically or manually.
In the definition module, the method specifically includes:
defining a priori information item and selecting an energy function item, taking a region-based energy function as an example, the prior information item Hspi(Ψ) we define as:
Hspi(Ψ)＝∫D|Ψ(x)-Ψspi(x)|2dx, (2)
therein ΨspiLevel set function representing the result of the prior information segmentation, Ψ represents the level set function being evolved, and for the energy function term, we choose to take a region-based energy function as an example, and the energy function is expressed in the form of an inner product:
Figure BDA0002631206870000101
s is the image data fit term, ω is a boundary detection operator defined as:
Figure BDA0002631206870000102
in the formula, x and y represent the positions of pixel points in the image, and lambda1,λ2Beta are all parameters greater than zero, GσIs a Gaussian kernel function, σ is the standard deviation, v1,v2Respectively, the approximate image intensities of the inside and outside of the active contour.
In the minimization module, the method specifically includes:
according to the formulas (2) and (3), the minimization problem of the model is as follows:
Figure BDA0002631206870000103
to solve this minimization problem, first, the Bregman variable b ═ b (b) is introducedx,by) And an auxiliary variable h ═ h (h)x,hy) So that
Figure BDA0002631206870000104
Then minimization problem equation (5) can be solved by the following equation:
Figure BDA0002631206870000105
wherein psik+1Is a level set function, hk+1Is an auxiliary variable, bk+1And bkIs the Bregman variable;
finally, we denote the evolving activity profile by z, defined as:
zk+1＝{x:Ψk+1(x)＝0}. (7)
the invention has the beneficial effects that: 1. in addition, the prior information automatically acquired by the active contour segmentation method can effectively reduce manual operation and improve the working efficiency; 2. because the general energy function is obtained from various existing mature algorithms, the active contour segmentation method of the invention not only increases the applicability to different types of images, but also increases the output of segmentation results; 3. the active contour segmentation method of the invention subsequently uses a split Bregman method to solve the minimized energy functional, and compared with the existing gradient descent method, the algorithm is robust to initialization. 4. The active contour segmentation method uses the convex function as the energy function, so that the local optimal solution can not be trapped in the process of minimizing the energy function, the segmentation result is accurate, and meanwhile, under the finite iteration, the speed of minimizing the energy functional by using the split Bregman method is higher than that of the traditional gradient descent method, so that the efficiency of the active contour segmentation method is greatly improved.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.

Claims (10)

1. An active contour segmentation method combining a general energy function and a priori information items, which is characterized by comprising the following steps:
step 1: selecting an acquisition mode of prior information to obtain a segmentation result of the fuzzy clustering as input psispi(ii) a Selecting an energy function as Hgen(Ψ)；
Step 2: inputting an image I to be segmented, and an initial level set function psi0Stop condition gamma, initial Bregman variable b00 and initial auxiliary variable h0＝0；
And step 3: calculating the approximate intensity of the image inside and outside the initial contour line
Figure FDA0002631206860000011
And 4, step 4: calculating an image data fitting term Sk；
And 5: computing the level set function Ψk+1；
Step 6: calculating an auxiliary vector hk+1And Bregman variable bk+1；
And 7: iteration
Figure FDA0002631206860000012
And 8: judge | | | Ψk+1-ΨkLess than or equal to gamma; if | | | Ψk+1-ΨkIf | | ≦ γ, executing step 9, otherwise returning to step 4, and continuing to calculate image data fitting item Sk；
And step 9: outputting the final result, i.e. the segmentation profile Zk+1。
2. The active contour segmentation method according to claim 1, further comprising, in the step 1, performing the steps of:
step S1: establishing a general image segmentation energy function based on prior information;
step S2, defining a prior information item and selecting an energy function item;
and step S3, selecting specific prior information to minimize the energy functional.
3. The active contour segmentation method according to claim 2, wherein in the step S1, the method specifically includes:
consider an image I:
Figure FDA0002631206860000014
Ψ:
Figure FDA0002631206860000015
is the level set function over the domain D, we define an energy function H (ψ):
Figure FDA0002631206860000013
wherein Hgen(Ψ) refers to a general energy function, and we select a boundary-based energy function, a region-based energy function, or other types of energy according to the characteristics and requirements of different imagesFunction, Hspi(psi) refers to a general a priori information item.
4. The active contour segmentation method according to claim 2, wherein in the step S2, the method specifically includes:
defining a priori information item and selecting an energy function item, the prior information item Hspi(Ψ) we define as:
Hspi(Ψ)＝∫D|Ψ(x)-Ψspi(x)|2dx, (2)
therein ΨspiLevel set function representing the result of the prior information segmentation, Ψ represents the level set function being evolved, and for the energy function term, we choose to take a region-based energy function as an example, and the energy function is expressed in the form of an inner product:
Figure FDA0002631206860000021
s is the image data fit term, ω is a boundary detection operator defined as:
Figure FDA0002631206860000022
in the formula, x and y represent the positions of pixel points in the image, and lambda1,λ2Beta are all parameters greater than zero, GσIs a Gaussian kernel function, σ is the standard deviation, v1,v2Respectively, the approximate image intensities of the inside and outside of the active contour.
5. The active contour segmentation method according to claim 2, wherein in the step S3, the method specifically includes:
according to the formulas (2) and (3), the minimization problem of the model is as follows:
Figure FDA0002631206860000023
to solve this minimization problem, first, the Bregman variable b ═ b (b) is introducedx,by) And an auxiliary variable h ═ h (h)x,hy) So that
Figure FDA0002631206860000025
Then minimization problem equation (5) is solved by the following equation:
Figure FDA0002631206860000024
wherein psik+1Is a level set function, hk+1Is an auxiliary variable, bk+1And bkIs the Bregman variable;
finally, we denote the evolving activity profile by z, defined as:
zk+1＝{x:Ψk+1(x)＝0}. (7)
6. an active contour segmentation system combining a generic energy function with a priori information items, comprising:
a selection unit: selecting the acquisition mode of prior information to obtain the segmentation result of fuzzy clustering as input psispi(ii) a Selecting an energy function as Hgen(Ψ)；
An input unit: for inputting an image I to be segmented, an initial level set function Ψ0Stop condition gamma, initial Bregman variable b00 and initial auxiliary variable h0＝0；
The first calculation unit: for calculating approximate intensities of images inside and outside initial contour lines
Figure FDA0002631206860000031
A second calculation unit: for calculating the fitting term S of image datak；
A third calculation unit: for computing level set functionsΨk+1；
A fourth calculation unit: for calculating auxiliary vectors hk+1And Bregman variable bk+1；
An iteration unit: for iteration
Figure FDA0002631206860000032
A judging unit: for judging | | Ψk+1-ΨkLess than or equal to gamma; if | | | Ψk+1-ΨkIf | | ≦ γ, executing step 9, otherwise returning to step 4, and continuing to calculate image data fitting item Sk；
An output unit: for outputting the final result segmentation contour Zk+1。
7. The active contour segmentation system according to claim 6, further comprising, in the selection unit:
a building module: the image segmentation energy function is used for establishing a general image segmentation energy function based on prior information;
a definition module: for defining a priori information term and selecting an energy function term;
a minimization module: the method is used for selecting specific prior information and minimizing the energy functional.
8. The active contour segmentation system according to claim 7, wherein the building block specifically includes:
consider an image I:
Figure FDA0002631206860000033
Ψ:
Figure FDA0002631206860000034
is the level set function over the domain D, we define an energy function H (ψ):
Figure FDA0002631206860000041
wherein Hgen(Ψ) refers to a general energy function, and we select a boundary-based energy function, a region-based energy function or other type of energy function, H, based on the characteristics and needs of the different imagesspi(psi) refers to a general a priori information item.
9. The active contour segmentation system according to claim 7, wherein the definition module specifically includes:
defining a priori information item and selecting an energy function item, the prior information item Hspi(Ψ) we define as:
Hspi(Ψ)＝∫D|Ψ(x)-Ψspi(x)|2dx, (2)
therein ΨspiLevel set function representing the result of the prior information segmentation, Ψ represents the level set function being evolved, and for the energy function term, we choose to take a region-based energy function as an example, and the energy function is expressed in the form of an inner product:
Figure FDA0002631206860000042
s is the image data fit term, ω is a boundary detection operator defined as:
Figure FDA0002631206860000043
in the formula, x and y represent the positions of pixel points in the image, and lambda1,λ2Beta are all parameters greater than zero, GσIs a Gaussian kernel function, σ is the standard deviation, v1,v2Respectively, the approximate image intensities of the inside and outside of the active contour.
10. The active contour segmentation system according to claim 7, characterized in that in the minimization module, it comprises in particular:
according to the formulas (2) and (3), the minimization problem of the model is as follows:
Figure FDA0002631206860000044
to solve this minimization problem, first, the Bregman variable b ═ b (b) is introducedx,by) And an auxiliary variable h ═ h (h)x,hy) So that
Figure FDA0002631206860000046
Then minimization problem equation (5) is solved by the following equation:
Figure FDA0002631206860000045
wherein psik+1Is a level set function, hk+1Is an auxiliary variable, bk+1And bkIs the Bregman variable;
finally, we denote the evolving activity profile by z, defined as:
zk+1＝{x:Ψk+1(x)＝0}. (7)
CN202010811710.8A 2020-08-13 2020-08-13 Active contour segmentation method and system combining general energy function and priori information item Active CN112102243B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010811710.8A CN112102243B (en) 2020-08-13 2020-08-13 Active contour segmentation method and system combining general energy function and priori information item

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010811710.8A CN112102243B (en) 2020-08-13 2020-08-13 Active contour segmentation method and system combining general energy function and priori information item

Publications (2)

Publication Number Publication Date
CN112102243A true CN112102243A (en) 2020-12-18
CN112102243B CN112102243B (en) 2023-06-09

Family

ID=73753625

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010811710.8A Active CN112102243B (en) 2020-08-13 2020-08-13 Active contour segmentation method and system combining general energy function and priori information item

Country Status (1)

Country Link
CN (1) CN112102243B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106447688A (en) * 2016-03-31 2017-02-22 大连海事大学 Method for effectively segmenting hyperspectral oil-spill image
CN108898611A (en) * 2018-06-19 2018-11-27 东华理工大学 Fuzzy region active contour parted pattern based on significant perception priori
CN109191477A (en) * 2018-09-21 2019-01-11 东华理工大学 Fuzzy region type active contour parted pattern based on overall situation and partial situation's fitting energy
CN110120057A (en) * 2019-04-16 2019-08-13 东华理工大学 Fuzzy region sexuality contours segmentation model based on the global and local fitting energy of weight
CN110969635A (en) * 2019-12-12 2020-04-07 哈尔滨工业大学（深圳） Robust rapid image segmentation method based on prior constraint level set framework
CN111462145A (en) * 2020-04-01 2020-07-28 重庆大学 Active contour image segmentation method based on double-weight symbol pressure function

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106447688A (en) * 2016-03-31 2017-02-22 大连海事大学 Method for effectively segmenting hyperspectral oil-spill image
CN108898611A (en) * 2018-06-19 2018-11-27 东华理工大学 Fuzzy region active contour parted pattern based on significant perception priori
CN109191477A (en) * 2018-09-21 2019-01-11 东华理工大学 Fuzzy region type active contour parted pattern based on overall situation and partial situation's fitting energy
CN110120057A (en) * 2019-04-16 2019-08-13 东华理工大学 Fuzzy region sexuality contours segmentation model based on the global and local fitting energy of weight
CN110969635A (en) * 2019-12-12 2020-04-07 哈尔滨工业大学（深圳） Robust rapid image segmentation method based on prior constraint level set framework
CN111462145A (en) * 2020-04-01 2020-07-28 重庆大学 Active contour image segmentation method based on double-weight symbol pressure function

Also Published As

Publication number Publication date
CN112102243B (en) 2023-06-09

Similar Documents

Publication Publication Date Title
Tang et al. Construction and verification of retinal vessel segmentation algorithm for color fundus image under BP neural network model
Dietenbeck et al. CREASEG: a free software for the evaluation of image segmentation algorithms based on level-set
CN107644420B (en) Blood vessel image segmentation method based on centerline extraction and nuclear magnetic resonance imaging system
CN104732213B (en) A kind of area of computer aided Mass detection method based on mammary gland magnetic resonance image
US8335359B2 (en) Systems, apparatus and processes for automated medical image segmentation
EP2936430B1 (en) Quantitative imaging
CN107292835B (en) Method and device for automatically vectorizing retinal blood vessels of fundus image
Díaz-Pernil et al. Fully automatized parallel segmentation of the optic disc in retinal fundus images
CN113012173A (en) Heart segmentation model and pathology classification model training, heart segmentation and pathology classification method and device based on cardiac MRI
CN104732520A (en) Cardio-thoracic ratio measuring algorithm and system for chest digital image
CN110969635B (en) Robust rapid image segmentation method based on priori constraint level set frame
Jiang et al. Level set based hippocampus segmentation in MR images with improved initialization using region growing
US7747051B2 (en) Distance transform based vessel detection for nodule segmentation and analysis
Jian et al. Diagnosis of left ventricular hypertrophy using convolutional neural network
Schenk et al. Automatic high-speed video glottis segmentation using salient regions and 3D geodesic active contours
Shabani et al. An active contour model using matched filter and Hessian matrix for retinalvessels segmentation
CN113658207B (en) Retina blood vessel segmentation method and device based on guided filtering
Padmasini et al. State-of-the-art of level-set methods in segmentation and registration of spectral domain optical coherence tomographic retinal images
CN110033861A (en) Suitable for the blood vessel of OCTA image and the quantitative analysis method of macula lutea avascular area and system
Zhao et al. Early diagnosis of cirrhosis via automatic location and geometric description of liver capsule
CN110648333B (en) Real-time segmentation system of mammary gland ultrasonic video image based on middle-intelligence theory
CN112102243B (en) Active contour segmentation method and system combining general energy function and priori information item
CN114926486B (en) Thyroid ultrasound image intelligent segmentation method based on multi-level improvement
Liu et al. Vct-net: An octa retinal vessel segmentation network based on convolution and transformer
CN108717699A (en) A kind of ultrasonography dividing method based on continuous minimal cut

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant