CN112330698B - Improved image segmentation method for geometric active contour - Google Patents
Improved image segmentation method for geometric active contour Download PDFInfo
- Publication number
- CN112330698B CN112330698B CN202011214403.8A CN202011214403A CN112330698B CN 112330698 B CN112330698 B CN 112330698B CN 202011214403 A CN202011214403 A CN 202011214403A CN 112330698 B CN112330698 B CN 112330698B
- Authority
- CN
- China
- Prior art keywords
- vector field
- level set
- advection
- active contour
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
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
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- 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
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an improved image segmentation method of a geometric active contour, which comprises the following steps of 1: inputting an image to be processed and a vector field; step 2: sampling a vector field through a rectangular grid, obtaining an observation set after sampling, and embedding an iterative robust estimator to eliminate errors and noises of an observed value in the observation set; then inserting observation points, and constructing an advection vector field by utilizing the smoothed ridge regression and constraining the elastic net; step 3: embedding the advection vector field and the diffusion flow into the geometric active contour, so as to construct a uniform geometric active contour model based on advection vector field and diffusion flow improvement to guide the updating of the initial level set, and the updated level set evolves an active contour curve to an accurate target contour, thereby completing the segmentation of the image. Compared with the traditional scheme, the method provided by the invention has obvious advantages and obtains a better segmentation effect.
Description
Technical Field
The invention relates to an improved image segmentation method of a geometric active contour, and belongs to the field of image processing.
Background
The contour captured from the image sequence has a rich spatial structure, and has been widely applied to video monitoring, medical analysis, motion recognition and the like. Features of interest in an image, whether rigid or non-rigid, typically involve complex motions and deformations. To accomplish this challenging task, an Active Contour (AC) method initiates an evolutionary curve near the object boundary, then matches the evolutionary curve to the true contour, and finally resides on the contour. Geometric Active Contour (GAC) captures the position of the parametric contour by minimizing the combination of smooth and gradient driven energies.
The curve evolution of the GAC model is essentially a Diffusion, e.g., a level set Function, considered as a spatial scalar model, and the non-uniform Diffusion is constrained by a Diffusion Coefficient Function (DCF) driven by the spatial data of the image. Many DCFs have been proposed to obtain the desired curve evolution behavior. The DCF of the original GAC is proposed based on a gradient-based stopping function that leads to curve shrinkage and boundary residuals. DCF has detailed spatial resolution, but the results are still sensitive to the initial level set, since the limited coordination range of DCF usually leads to local convergence. When dynamic segmentation of time series is involved, the disadvantages of DCF are very prominent, often confusing the true contours and the salient background.
Many models assume that evolving contour-aligned moving objects are time-smooth in the sequence, so the evolving contour at the current time can be initialized by the corresponding contour at the previous time. However, considering the sampling interval, even if diffusion by gaussian smoothing has been previously achieved to the original image, the global motion between neighboring images may be so significant that the DCF is discontinuous in time. To overcome this problem, predictive methods are embedded into the GAC model to initialize the level set, otherwise local convergence per discrete time will accumulate, eventually losing object boundaries.
Disclosure of Invention
The invention provides an improved image segmentation method of a geometric active contour, which is used for realizing the segmentation of an image by constructing the improved geometric active contour.
The technical scheme of the invention is as follows: an improved image segmentation method of geometric active contour, the method steps are as follows:
step 1: inputting an image to be processed and a vector field;
step 2: sampling a vector field through a rectangular grid, obtaining an observation set after sampling, and embedding an iterative robust estimator to eliminate errors and noises of an observed value in the observation set; then inserting observation points, and constructing an advection vector field by utilizing the smoothed ridge regression and constraining the elastic net;
Step 3: embedding the advection vector field and the diffusion flow into a geometric active contour, so as to construct a uniform geometric active contour model based on advection vector field and diffusion flow improvement to guide the updating of an initial level set, and evolving an active contour curve to an accurate target contour by the updated level set, thereby completing the segmentation of the image; wherein an internal contour of an image is captured from a sequence of images by means of a conventional geometric active contour model and an initial level set is generated.
Step2, establishing an energy equation E of a vector field according to a Lagrange method and concisely writing the energy equation E into a matrix form:
in the formula:is a column vector ordering a conventional grid matrix in dictionary order, L is a matrix smoothing each element of the vector field X by the Laplacian, D is 4 grid points near the observation point of the matrix insertion, pointing to its own position in a bilinear manner, λ 1 And λ 2 Is to satisfy lambda 1 +λ 2 A positive coefficient of 1;
obtaining a regular equation by solving partial derivatives in an energy equationAnd solving by using ridge regression to obtain advection vector fieldComprises the following steps:
wherein U ═ D T D+λ 1 L T L) -1 D T 。
The Step3 is specifically as follows:
let the evolution of the curve be unlimited, the level set equation can be generalized by transferring spatial intensity through two processes inside the closed system: diffusion and advection; the geometric active contour model combines the divergence of diffusion vectors and advection vectors and can be decomposed into a divergence operation of the product of scalar and vector functions:
In the formula, the level set phi represents the image spaceThe variable of interest, g represents a function mapping the image intensity matrix to the corresponding diffusion coefficient, and the vector field X represents a finite vector spaceThe advection velocity in (1); on the right side of the equation, the first termDenotes an unrestricted diffusion of phi; second itemRepresenting a motion driven by the edges and the gradient of the advection vector field, which motion is activated if the intensity is significant; the third item retains the heterogeneityConservation of uniform force field; the last term compensates for the change in amount;
the level set formula is expanded to be embedded in the tangential direction of the flat flow field, and the tangential direction isThe divergence operation of (a) can also be decomposed into:
when a singular point having a large curvature is present near the vector field X,represents the curvature of the vector field X and is typically small;
the curvature compensation C is used to enhance the detection performance of the evolution curve C driven by the level set:
in addition, the gradient direction of the curve CAt a speed value ofTangential direction of the vector field XAt a speed value of
Then, the velocity vector is projected to the gradient directionThe evolution equation for the level set function is formulated as:
the updated level set continues to guide the curve evolution of the geometric active contour to capture a dynamic contour with accurate complex motion and complete the segmentation of the image.
The beneficial effects of the invention are: the invention provides a regression algorithm to construct an advection vector field, and utilizes the advection vector field and a diffusion flow to guide the update of a level set, an initial level set generated by a geometric active contour evolves towards an accurate dynamic contour under the guide of the advection vector field and the diffusion flow, the image segmentation is realized, the dynamic shape of the contour can be more accurately captured through the updated level set, and the problem that the traditional active contour model cannot accurately segment the dynamic complex image is solved; experimental results show that compared with the traditional scheme, the method provided by the invention has obvious advantages and obtains a better segmentation effect.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a graph showing the results of the experiment according to the present invention.
Detailed Description
Example 1: 1-2, an improved method for image segmentation of geometric active contours, the method comprising the steps of:
step 1: inputting an image to be processed and a vector field;
step 2: sampling a vector field through a rectangular grid, obtaining an observation set after sampling, and embedding an iterative robust estimator to eliminate errors and noises of an observed value in the observation set; then inserting observation points, and constructing an advection vector field by utilizing the smoothed ridge regression and constraining the elastic net;
Step 3: embedding the advection vector field and the diffusion flow into a geometric active contour, so as to construct a uniform geometric active contour model based on advection vector field and diffusion flow improvement to guide the updating of an initial level set, and evolving an active contour curve to an accurate target contour by the updated level set, thereby completing the segmentation of the image; wherein an internal contour of an image is captured from a sequence of images by means of a conventional geometric active contour model and an initial level set is generated.
Further, Step1 can be set to refer to inputting the image to be processed, and simultaneously inputting a vector field X:
further, Step2 may be specifically configured as follows:
first, with a rectangular grid P of length-width dimension H, W ═ H × W versus continuous vector field X:sampling is carried out; sampling to obtain an observation setAnd N is less than P, an iterative robust estimator is embedded to eliminate errors and noises of observed values in an observation set; and obtaining the advection vector field by realizing the ridge regression after smoothing and modifying the elastic net. Because of the adverse effects of the smoothing term, ridge regression tends to activate vectors for almost every location. In order to obtain accurate and significant results, the energy equation of the vector field is established according to the Lagrangian method and concisely written in the form of a matrix:
WhereinIs a column vector ordering a conventional grid matrix X in dictionary order, L is a matrix smoothing each element of X by the Laplacian, and D is a matrix inserting 4 grid points near the observation point pointing to its own position in a bilinear manner, λ 1 And λ 2 Is to satisfy lambda 1 +λ 2 A positive coefficient of 1. Obtaining a regular equation by solving partial derivatives in the equationAnd solving by using ridge regression to obtain advection vector fieldComprises the following steps:
U=(D T D+λ 1 L T L) -1 D T
and finally, embedding the constructed advection vector field into a solution of a subsequent step.
Further, Step4 can be set, and the specific steps of constructing a unified geometric active contour model (AD-GAC) based on advection vector field and diffusion flow improvement are as follows:
let the evolution of the curve be unlimited, the level set equation can be generalized by transferring spatial intensity through two processes inside the closed system: diffusion and advection. AD-GAC combines the divergence of diffusion and advection vectors and can be decomposed into a divergence operation of the product of a scalar and a vector function:
in the formula, the level set phi represents the image spaceThe variable of interest, g represents a function mapping the image intensity matrix to the corresponding diffusion coefficient, and the vector field X represents a finite vector space The advection velocity in (2). On the right side of the equation, the first termDenotes an unrestricted diffusion of phi; second itemRepresenting a motion driven by the edges and the gradient of the advection vector field, which motion is activated if the intensity is significant; the third term preserves the conservation of the non-uniform force field; the last term compensates for the change in the amount.
The level set formula is expanded to be embedded in the tangential direction of the flat flow field, and the tangential direction isThe divergence operation of (d) can also be decomposed into:
when a singular point having a large curvature is present near the vector field X,represents the curvature of the vector field X and is typically small.
The curvature compensation C is used to enhance the detection performance of the evolution curve C driven by the level set:
in addition, the method can be used for producing a composite materialGradient direction of curve CAt a speed value ofTangential direction of the vector field XAt a speed value of
Then, the velocity vector is projected to the gradient directionThe evolution equation of the function of the level set is formulated as
Wherein s represents a set tuning parameter;
the updated level set continues to guide the curve evolution of the geometric active contour to capture a dynamic contour with accurate complex motion and complete the segmentation of the image.
By the method of the invention, the following experimental data are given:
As shown in fig. 2, the microbial sequence indicates that the microbes perform spontaneous activities in the nutrient solution, which involves overall rotation and detailed deformation. To initiate the segmentation, 3 individual sub-regions in the body of the microorganism are manually related to the representation. The light gray circle represents an initial frame of the current moment t, the initial frame moves towards the segmentation contour, the dark gray circle represents a final segmentation result, and the initial frame is closer to the final segmentation result under the action of an advection vector field and a diffusion flow along with the increase of t.
In the experimental process of the invention, a system win10 is used, and a computer configured as an AMD R52600 processor, a 16G running memory and a GeForce RTX 1070(8GB) graphics card is adopted.
The working principle of the invention is as follows: in the invention, a uniform GAC model, namely an improved geometric active contour model (AD-GAC) based on advection vector field and diffusion flow, and an embedded algorithm for advection vector field regression between adjacent frames are provided. To segment dynamic objects between image sequences, the AD-GAC method extends the conventional diffusion scheme according to the generation advection diffusion equation. The advection term of the advection diffusion equation is represented by the transfer of the level set function of the time-varying vector field. Furthermore, a ridge regression of the constrained norm is implemented and the elastic mesh is modified to obtain the vector field. In addition, errors and noise of the observed values are eliminated by embedding an iterative robust estimator. The experimental results prove that the method of the invention is superior to the traditional method based on the Active Contour (AC), and especially has more advantages under the condition of obvious global motion.
While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.
Claims (2)
1. An improved image segmentation method of geometric motion contour is characterized in that: the method comprises the following steps:
step 1: inputting an image to be processed and a vector field;
step 2: sampling the vector field through a rectangular grid, obtaining an observation set after sampling, and embedding an iterative robust estimator to eliminate errors and noises of an observed value in the observation set; then inserting observation points, and constructing an advection vector field by utilizing the smoothed ridge regression and constraining the elastic net;
step 3: embedding the advection vector field and the diffusion flow into a geometric active contour, so as to construct a uniform geometric active contour model based on advection vector field and diffusion flow improvement to guide the updating of an initial level set, and evolving an active contour curve to an accurate target contour by the updated level set, thereby completing the segmentation of the image; wherein an internal contour of an image is captured from a sequence of images by means of a conventional geometric active contour model and an initial level set is generated;
The Step3 is specifically as follows:
let the evolution of the curve be unlimited, the level set equation can be generalized by transferring spatial intensity through two processes inside the closed system: diffusion and advection; the geometric active contour model combines the divergence of diffusion vectors and advection vectors and can be decomposed into a divergence operation of the product of scalar and vector functions:
in the formula, the level set phi represents the image spaceThe variable of interest, g represents a function mapping the image intensity matrix to the corresponding diffusion coefficient, and the vector field X represents a finite vector spaceThe advection velocity in (1); on the right side of the equation, the first termDenotes an unrestricted diffusion of phi; second itemRepresenting a motion driven by the edges and the gradient of the advection vector field, which motion is activated if the intensity is significant; the third term preserves the conservation of the non-uniform force field; the last term compensates for the change in quantity;
the level set formula is expanded to be embedded in the tangential direction of the flat flow field, and the tangential direction isThe divergence operation of (a) can also be decomposed into:
when a singular point having a large curvature is present near the vector field X,represents the curvature of the vector field X and is typically small;
the curvature compensation C is used to enhance the detection performance of the evolution curve C driven by the level set:
In addition, the gradient direction of the curve CAt a speed value ofTangential direction of the vector field XAt a speed value of
Then, the velocity vector is projected to the gradient directionThe evolution equation for the level set function is formulated as:
the updated level set continues to guide the curve evolution of the geometric active contour to capture a dynamic contour with accurate complex motion and complete the segmentation of the image.
2. The improved image segmentation method of geometric active contour according to claim 1, characterized in that: step2, establishing an energy equation E of a vector field according to a Lagrange method and concisely writing the energy equation E into a matrix form:
in the formula:is a column vector ordering the conventional grid matrix in dictionary order, L is the vector field X by Laplace's operator for each elementA matrix subjected to smoothing processing, D is a matrix insertion observation point b i Nearby 4 grid points pointing in a bilinear manner to their own position, λ 1 And λ 2 Is to satisfy lambda 1 +λ 2 A positive coefficient of 1;
obtaining a regular equation by solving partial derivatives in an energy equationAnd solving by using ridge regression to obtain advection vector fieldComprises the following steps:
Wherein U is (D) T D+λ 1 L T L) -1 D T 。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011214403.8A CN112330698B (en) | 2020-11-04 | 2020-11-04 | Improved image segmentation method for geometric active contour |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011214403.8A CN112330698B (en) | 2020-11-04 | 2020-11-04 | Improved image segmentation method for geometric active contour |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112330698A CN112330698A (en) | 2021-02-05 |
CN112330698B true CN112330698B (en) | 2022-07-29 |
Family
ID=74324681
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011214403.8A Active CN112330698B (en) | 2020-11-04 | 2020-11-04 | Improved image segmentation method for geometric active contour |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112330698B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117253228B (en) * | 2023-11-14 | 2024-02-09 | 山东大学 | Cell cluster space constraint method and system based on nuclear image distance intra-coding |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109509202A (en) * | 2018-09-21 | 2019-03-22 | 浙江大学 | A kind of batterylaying image partition method based on improvement movable contour model |
CN109636816A (en) * | 2018-11-21 | 2019-04-16 | 中国电子科技集团公司第二十八研究所 | A kind of ultrasonic image division method |
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 |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1938271A2 (en) * | 2005-10-21 | 2008-07-02 | The General Hospital Corporation | Methods and apparatus for segmentation and reconstruction for endovascular and endoluminal anatomical structures |
US10074208B2 (en) * | 2015-08-21 | 2018-09-11 | The Regents Of The University Of California | Automated comparison of 3D images |
CN109087309A (en) * | 2018-07-19 | 2018-12-25 | 华南理工大学 | A kind of image partition method of amalgamation of global and local information level collection |
-
2020
- 2020-11-04 CN CN202011214403.8A patent/CN112330698B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109509202A (en) * | 2018-09-21 | 2019-03-22 | 浙江大学 | A kind of batterylaying image partition method based on improvement movable contour model |
CN109636816A (en) * | 2018-11-21 | 2019-04-16 | 中国电子科技集团公司第二十八研究所 | A kind of ultrasonic image division method |
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 |
Non-Patent Citations (5)
Title |
---|
Honggang Yu 等.Robust Segmentation of Freehand Ultrasound Image Slices Using Gradient Vector Flow Fast Geometric Active Contours.《2006 IEEE Southwest Symposium on Image Analysis and Interpretation》.2006,第115-119页. * |
K. Mikula 等.New fast and stable Lagrangean method for image segmentation.《2012 5th International Congress on Image and Signal Processing》.2013,第688-696页. * |
KhamaelAL-Dulaimi 等.Evaluation and benchmarking of level set-based three forces via geometric active contours for segmentation of white blood cell nuclei shape.《Computers in Biology and Medicine》.2020,第116卷第1-15页. * |
欧玥 等.改进测地线活动轮廓模型的肝脏超声图像分割.《中国测试技术》.2008,第34卷(第3期),20080530. * |
赵文成.聚焦超声治疗肿瘤的超声图像轮廓提取方法研究.《中国优秀博硕士学位论文全文数据库(博士) 医药卫生科技辑》.2019,E060-4. * |
Also Published As
Publication number | Publication date |
---|---|
CN112330698A (en) | 2021-02-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7623731B2 (en) | Direct method for modeling non-rigid motion with thin plate spline transformation | |
Niemeyer et al. | Occupancy flow: 4d reconstruction by learning particle dynamics | |
JP7433297B2 (en) | Deep learning-based coregistration | |
CN103854262B (en) | Medical image denoising method based on documents structured Cluster with sparse dictionary study | |
CN111080675B (en) | Target tracking method based on space-time constraint correlation filtering | |
US20230169727A1 (en) | Generative Nonlinear Human Shape Models | |
CN111814839B (en) | Template matching method of longicorn group optimization algorithm based on self-adaptive variation | |
JP2010238226A (en) | Method and system for tracking object | |
Brunet et al. | Feature-driven direct non-rigid image registration | |
CN112330698B (en) | Improved image segmentation method for geometric active contour | |
Chan et al. | Two and three dimensional image registration based on B-spline composition and level sets | |
Fortun et al. | Fast piecewise-affine motion estimation without segmentation | |
López-Tapia et al. | Fast and robust cascade model for multiple degradation single image super-resolution | |
CN114387392A (en) | Method for reconstructing three-dimensional human body posture according to human shadow | |
CN112184549B (en) | Super-resolution image reconstruction method based on space-time transformation technology | |
JP6806160B2 (en) | 3D motion evaluation device, 3D motion evaluation method, and program | |
Ebrahimi et al. | Multi-Frame Super-Resolution with No Explicit Motion Estimation. | |
Yan et al. | Weighted sparse coding residual minimization for visual tracking | |
Lee et al. | Carrying object detection using pose preserving dynamic shape models | |
CN110956700A (en) | Density regulation and control method for generating point cloud based on motion recovery structure | |
Gay-Bellile et al. | Feature-driven direct non-rigid image registration | |
Ruijters et al. | Efficient GPU-accelerated elastic image registration | |
Liang et al. | Spatiotemporal super-resolution reconstruction based on robust optical flow and Zernike moment for video sequences | |
CN106204445A (en) | Image/video super-resolution method based on structure tensor total variation | |
Zhu et al. | Efficient illumination insensitive object tracking by normalized gradient matching |
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 |