CN110473224B - Automatic RSF level set image segmentation method based on KL entropy - Google Patents

Automatic RSF level set image segmentation method based on KL entropy Download PDF

Info

Publication number
CN110473224B
CN110473224B CN201910764486.9A CN201910764486A CN110473224B CN 110473224 B CN110473224 B CN 110473224B CN 201910764486 A CN201910764486 A CN 201910764486A CN 110473224 B CN110473224 B CN 110473224B
Authority
CN
China
Prior art keywords
image
entropy
rsf
segmented
level set
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
Application number
CN201910764486.9A
Other languages
Chinese (zh)
Other versions
CN110473224A (en
Inventor
邹乐
章义刚
王晓峰
周琼
龙夏
张惯虹
邓锐
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.)
Hefei University
Original Assignee
Hefei University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hefei University filed Critical Hefei University
Priority to CN201910764486.9A priority Critical patent/CN110473224B/en
Publication of CN110473224A publication Critical patent/CN110473224A/en
Application granted granted Critical
Publication of CN110473224B publication Critical patent/CN110473224B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20092—Interactive image processing based on input by user
    • G06T2207/20104—Interactive definition of region of interest [ROI]
    • 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
    • G06T2207/30016—Brain

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Operations Research (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an RSF level set image automatic segmentation method based on KL entropy, which comprises the following steps: acquiring an image to be segmented; carrying out significance analysis on an image to be segmented, and determining an initial level set of an RSF model; calculating KL entropy of the image, taking KL entropy values of the image as weight coefficients of internal and external energy of a segmentation curve, introducing the KL entropy of the image into a level set function of an RSF model, and determining a fitting energy functional of the image to be segmented; solving an evolution equation according to the fitting energy functional of the image to be segmented, and iteratively solving according to the evolution equation to determine an image segmentation result. The image edge detection method based on the new RSF model is used for improving the detection capability of the image edge, namely KL entropy is used as a weight coefficient of internal and external energy of a curve, the internal energy of the model is local field energy near the curve, the image segmentation effect on the images with uneven gray scale and low contrast is better than that of the RSF model, and the same image is segmented under the condition of same parameter setting, so that the same segmentation result can be obtained by fewer iteration times.

Description

Automatic RSF level set image segmentation method based on KL entropy
Technical Field
The invention relates to the technical field of image segmentation, in particular to an RSF level set image automatic segmentation method based on KL entropy.
Background
In image research, people usually are interested in a certain area, and image segmentation can extract the interested part, and therefore, as a leading edge subject, the image research is full of challenges. Image segmentation has wide application in many fields, such as the aeronautical field, the medical field, geographic mapping, and so on. For example, in the medical field, images are one of the important bases for doctors to make diagnoses, and play a very important role, and segmentation of images enables doctors to obtain effective medical information. The effective basis for segmenting the image is the brightness and color of the pixels of the image. Without a correct segmentation, a correct recognition cannot be obtained. However, if the segmentation is performed only according to the brightness and the color of the pixels of the image, the segmentation will encounter many obstacles, such as uneven gray scale, uneven illumination, noise influence, shadow, etc., which often cause the segmentation of the image to be wrong. Although many methods for extracting edges and segmenting regions have been developed, no method has been developed which can be applied to all images, and therefore, introducing a new method to obtain a correct image segmentation result is a key point and difficulty in image processing research.
Image segmentation is a difficult problem in the field of computer vision, and countless researchers have been attracted to develop the image segmentation in the last 70 th century, so that a plurality of segmentation algorithms are proposed. In the first place, in 1962, Doyle proposed a P-tile algorithm, which is the oldest threshold segmentation method, and the algorithm has good noise resistance but has no power for images with difficult prior probability estimation. In 1978, the Dajin Zhan proposed the maximum inter-class variance method, which has a simple algorithm and can effectively cut images when the area of the background is close to that of the target. When the area difference between the two is large, the segmentation effect is not good. In 1985, kapcur et al proposed a one-dimensional maximum entropy thresholding method that also cuts well for non-ideal bi-modal histograms, but is computationally expensive. In 1989, Abutaleb popularized the method to two dimensions based on the one-dimensional maximum entropy threshold method. In recent years, researchers have proposed many new methods for image segmentation, such as global binarization and edge detection algorithms proposed by Zhao Xue pine, and the like. Image segmentation is a classic problem in the image field, each algorithm has its own unique features and its disadvantages, and the segmentation technology is still in continuous research and development.
The CV model (Chan-Vese model) is based on the level set of the region, and since no edge is involved, minimizing it can result in the boundary of the target object. Due to the effects of heterogeneity and complex configurations, the performance of level set segmentation can become apparent by the presence of nearby structures of similar intensity, making it impossible to identify the precise boundaries of objects. Furthermore, even with an optimal configuration of the control parameters, the CV model does not achieve very accurate segmentation results, thus requiring a significant amount of manual intervention.
In summary, the conventional CV model only uses gray homogeneity as a criterion for region separation, and can only be used for segmenting an image with high contrast between a target and a background, while the CV model has a poor segmentation effect for a complex and non-uniform image.
Disclosure of Invention
The embodiment of the invention provides an automatic RSF level set image segmentation method based on KL entropy, which is used for solving the problems in the background technology.
The embodiment of the invention provides an RSF level set image automatic segmentation method based on KL entropy, which comprises the following steps:
acquiring an image to be segmented;
carrying out significance analysis on an image to be segmented, and determining an initial level set of an RSF model;
calculating KL entropy of the image, taking KL entropy values of the image as weight coefficients of internal and external energy of a segmentation curve, introducing the KL entropy of the image into a level set function of an RSF model, and determining a fitting energy functional of the image to be segmented;
solving an evolution equation according to the fitting energy functional of the image to be segmented, and iteratively solving according to the evolution equation to determine an image segmentation result.
Further, the image to be segmented is subjected to significance analysis; the method specifically comprises the following steps: a saliency map is obtained using a residual spectrum method.
Further, KL entropy of the image; the method specifically comprises the following steps:
p i to p 0 KL entropy of (a) is expressed as follows:
Figure BDA0002171490830000031
p 0 to p i KL entropy of (a) is expressed as follows:
Figure BDA0002171490830000032
wherein x is a point on the image; p is a radical of i And p 0 Probability density distribution functions for the inner region and the outer region, respectively; φ is the level set function of the RSF model.
Further, fitting an energy functional of the image to be segmented; the method specifically comprises the following steps:
Figure BDA0002171490830000033
wherein KL (p) i /p 0 ) And KL (p) 0 /p i ) Internal and external energy weight coefficients; f. of 1 (x) And f 2 (x) The mean values of the image gray levels inside and outside the segmentation curve respectively; h (φ) represents a Heaviside function; i (y) is a given image; k σ Is a Gaussian function; upsilon and mu are positive weighting constants; omega 1 、Ω 2 And Ω is an integration area.
Further, the evolution equation specifically includes:
Figure BDA0002171490830000034
the embodiment of the invention provides an RSF level set image automatic segmentation method based on KL entropy, which has the following beneficial effects compared with the prior art:
the invention provides an RSF level set image automatic segmentation method based on KL entropy, which is used for improving the detection capability of image edges, acquiring an initial contour by using significance detection, and using Kullback-Leibler entropy as a weight coefficient of internal and external energy of a curve, wherein the internal energy of a model is local field energy near the curve, while the original RSF (region-scalable matching energy) model segmentation is seriously influenced by noise, uneven gray scale and low contrast. In particular, segmentation can be performed on some radar images, and the image segmentation effect on the influence of noise is better than that of the CV model.
Drawings
FIG. 1 is a flowchart of an RSF level set image automatic segmentation method based on KL entropy according to an embodiment of the present invention;
fig. 2 is an original graph of a picture with uneven gray distribution, an initial contour obtained by saliency detection, a KLRSF model segmentation result graph, and a CV model segmentation result graph provided by the embodiment of the present invention;
fig. 3 is a maple leaf original image with uneven gray scale, an initial contour obtained by saliency detection, a KLRSF model segmentation result image, and a CV model segmentation result image provided by the embodiment of the present invention;
fig. 4 is a brain original image, an initial contour obtained by saliency detection, a KLRSF model segmentation result image, and a CV model segmentation result image provided by the embodiment of the present invention;
fig. 5 is a noise original graph, an initial contour obtained by saliency detection, a KLRSF model segmentation result graph, and a CV model segmentation result graph provided by the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, an embodiment of the present invention provides a KL entropy-based RSF level set image automatic segmentation method, including:
step 1, obtaining an image to be segmented.
And 2, performing significance analysis on the image to be segmented, and determining an initial level set of the RSF model.
And 3, calculating KL entropy of the image, taking the KL entropy value of the image as a weight coefficient of external energy and internal energy of the segmentation curve, introducing the KL entropy of the image into a level set function of the RSF model, and determining a fitting energy functional of the image to be segmented.
And 4, solving an evolution equation according to the fitting energy functional of the image to be segmented, and iteratively solving according to the evolution equation to determine an image segmentation result.
For step 1, most algorithms for image segmentation are based on discontinuity of gray values and similar properties. In the former, segmentation algorithms segment an image based on the occurrence of a discontinuity in gray scale. Assuming that the boundaries of different regions of an image are completely different from each other and from the background, this allows detection of edges based on local discontinuities in the gray scale. Which segments the image into similar regions according to a set of predefined disciplines. The image segmentation algorithm mainly comprises the following steps: edge-based, threshold-based, region-based, cluster-analysis-based, wavelet-transform-based, mathematical morphology-based, artificial neural network-based. The first three of which are most common.
(1) Edge segmentation of the image: an edge is a collection of pixels with abrupt gray changes in an image, and is generally detected by differentiation, and the gray changes of the pixels along the edge are relatively smooth, and the gray changes of the pixels perpendicular to the edge are severe. According to the characteristics of gray scale change, the method can be divided into three types: stepped, roof-shaped, flange-shaped. The edge detection methods are various, and mainly include the following methods: the method comprises the following steps of spatial domain differential operator, fitting curved surface, wavelet multi-scale and mathematical morphology-based.
(2) Threshold segmentation of the image: the threshold segmentation has the advantages of only management, simple realization and high calculation speed. The thresholding is at the core defense in the segmentation application. The thresholding may be divided into single thresholding and multi-thresholding.
(3) Segmentation of the image: both region growing algorithms and region splitting aggregation are region-based segmentation algorithms. The region growing algorithm is a process of grouping pixels into larger regions according to previously defined criteria. The region splitting and aggregating is to divide a region into a plurality of different regions and then split and aggregate the regions according to certain requirements.
Generally, the boundary of the image to be segmented is very sharp, and it is enough to use the edge-based segmentation method, but in the case of medical images, such as heart images, brain images, etc., the edge-based algorithm cannot be used, because the boundary of the image is not clear, and the difference between each region is only the same color, but different shades, which is practical when using the region-based method.
For step 2, in the evolution of the active contour model, the selection of the initial contour is usually iterated from the edge of the whole image, so that the result of contour evolution is often interfered by background information, and a satisfactory target result cannot be obtained. Also, in time terms, if the initial contour can evolve from close proximity to the target object, the time it takes is reduced. Therefore, the introduction of the significance analysis of the image is considered, and the combination of the significance map enables the position of the target object, namely the approximate outline of the object, to be pre-judged, so that the interference of background information on the target is avoided, the number of times of outline evolution is greatly reduced, and the effect and the efficiency of the traditional active outline model are improved. Firstly, determining the approximate position of a target in a complex image background by using a saliency map of an image to obtain the initial position of the evolution of the active contour; meanwhile, evolution can be carried out only around the target object, so that the interference of a complex background on the evolution is avoided, the times of the outline evolution are greatly reduced, and the effect and the efficiency of the traditional active outline model are improved. In the invention, a residual spectrum method is used for obtaining a saliency map of an image.
For step 3, KL entropy of image; the method specifically comprises the following steps:
p i to p 0 KL entropy of (a) is expressed as follows:
Figure BDA0002171490830000061
p 0 to p i KL entropy of (a) is expressed as follows:
Figure BDA0002171490830000062
wherein x is a point on the image; p is a radical of i And p 0 Probability density distribution functions for the inner region and the outer region, respectively; φ is the level set function of the RSF model.
The earliest KL distance, also called relative entropy, was introduced from the information theory and measures the difference between two probability distributions in the same event space. The physical significance is as follows: in the same event space, the probability distribution p (x) corresponds to each event, and if the event is coded by the probability distribution q (x), the coding length of each event is increased by an average of a number of bits. In matching objects and tracking problems, the region of the matching object can be found by the smallest KL distance. Several uses of KL distance: measuring the difference of the two probability distributions; measuring the energy loss when the probability distribution P is fitted by using the probability distribution Q, namely, the amount of information lost after fitting; the similarity of the two probability distributions is measured, and the motion without the added notes and the motion with the added labels can be measured in the motion capture, so that the motion classification is carried out.
For step 3, the Region-Scalable fixing (RSF) model specifically includes:
the RSF model is proposed by chuning Li et al, which can well process uneven gray level images, and its basic idea is: introducing a local fitting function to express the gray value of a local region of a target boundary, controlling the size of the local region by using a Gaussian kernel (Gaussian kernel) function window, and defining a local energy functional by analyzing the modified average gray value of the local region of the image as follows:
Figure BDA0002171490830000071
wherein x is a point on the image; omega 1 ,Ω 2 Similar to CV model, show inside (C), outside (C); f. of 1 (x),f 2 (x) Representing gray values in different local regions of the image; k σ Which is a gaussian kernel function, for implementing image smoothing, can be expressed as follows:
Figure BDA0002171490830000072
fitting x in the model, calculating and solving an energy functional of the x, and introducing a level set function, wherein the expression is as follows:
Figure BDA0002171490830000081
the third item in the expression is a length penalty item and is used for constraining the length of the curve; the fourth term is a functional penalty to avoid re-initialization during evolution.
Then, the level set function is subjected to variational solution to obtain the following evolution equation:
Figure BDA0002171490830000082
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0002171490830000083
a weight parameter representing a length penalty term; mu is a regularization weight parameter, and is generally greater than or equal to 0.
For step 4, the weight parameters are difficult to set for the segmentation of the heterogeneous image. υ and μ are positive weighting constants. In the conventional RSF model, λ o And λ b Uniformity weights inside and outside the curve, respectively, do not provide a method to make a reasonable choice of these parameters. In the desired segmentationWhen the image with a constant structure is segmented, the requirements on the weight parameters are loose. The presence of small gray scale differences of the segmented object from the background, as well as the heterogeneity of the regions and similar intensity structures nearby, can lead to the detection of false boundaries. Therefore, the conventional RSF model cannot achieve accurate segmentation.
When the internal probability is greater than the external probability, and the internal uniformity is greater than the external uniformity, the weight of the external uniformity in the energy function needs to be increased to reduce the influence of smaller uniformity on energy; when the internal probability is less than the external probability, the weight of the internal uniformity must be enhanced. The energy function will be minimal when the homogeneity of the two regions is balanced.
To address the difficulties mentioned above, KL entropy is substituted for λ o and λ b . The fitting energy is expressed as:
Figure BDA0002171490830000091
along with the change of the KL entropy along with the evolution curve, the weight of the uniformity of the two regions in the function is automatically adjusted in a self-adaptive manner, and the lowest energy is obtained.
The curve evolution equation derived according to the Euler-Lagrange equation specifically comprises the following steps:
Figure BDA0002171490830000092
according to the technical scheme, the automatic image segmentation method of the RSF level set based on the KL entropy is provided for improving the detection capability of the image edge, Kullback-Leibler Divergence (KL entropy) is used as a weight coefficient of internal and external energy of a curve, the internal energy of a model is local domain energy near the curve, the segmentation of an original RSF (region-scalable shaping energy) model is seriously influenced by noise, uneven gray scale and low contrast, the segmentation effect of the newly improved KLRSF model (the RSF model based on the KL entropy) on the images with uneven gray scale and low contrast is better than that of the RSF model, the same image is segmented under the condition of same parameter setting, and the same segmentation result can be obtained by reducing the number of iterations.
Comparison of CV model and KLRSF model
CV model: the method has the advantages of low segmentation speed and efficiency, can only be used for segmenting target and background high-contrast images, cannot segment images with uneven gray levels, needs to initialize a level set function continuously in order to keep the characteristic of a symbol distance function in the curve evolution process, and is large in calculation amount and low in efficiency.
KLRSF model: the KLRSF model is improved on the basis of the CV model. The image with uneven gray scale can be divided, and the setting requirement on the parameters is relaxed.
Wherein, the CV model specifically includes:
i represents a given image on a defined field omega, C is a closed curve of an object and a background image, and omega is used in the curve i Is used to denote, conversely, Ω 0 The background of the representation is external. The energy functional is thus expressed as:
Figure BDA0002171490830000101
wherein L (C) represents the length of C of the curve, and S (C) represents Ω i Area of (d), mu>0,ν>0,λ a ,λ b >0 is a weight coefficient, and C1 and C2 are the image gray level means inside and outside the evolution curve C, respectively.
In the level-set approach, C is considered as a zero level set of φ, φ (Ω) i >0)，φ(Ω 0 <0) The energy functional formula is expressed in terms of phi as:
Figure BDA0002171490830000102
wherein H (φ) represents the Heaviside function and δ (φ) represents the Dirac function.
The variables c1 and c2 may be expressed as:
Figure BDA0002171490830000103
Figure BDA0002171490830000104
f (φ) is the fitting energy, which can be expressed as:
F(φ)＝λ a |I(z)-c1| 2 H(φ)+λ b |I(z)-c2| 2 (1-H(φ))
from the energy functional of the CV model, the curve evolution is mainly influenced by two main terms, the first is to regularize the contour so that the contour keeps smooth in the convolution process, and the second is the data term and has great influence on the evolution of the contour line. The model does not block the evolution curve on the object boundary based on edge factors, but makes the inside and outside of the curve have the best uniformity. The advantage of this model is robustness to noise. It is well known that it is suitable for image segmentation of two regions with a distinct average of pixel intensities.
Analysis of Experimental results
1. The picture with uneven gray distribution is divided, see fig. 2.
The first column is the image to be segmented, the second column is the image obtained by using SR saliency detection, the third column is the result obtained by the present invention, and the fourth column is the result obtained by the CV model. The comparison of experimental results shows that under the condition of the same parameters, the KLRSF model is successfully segmented, the CV model is unsuccessfully segmented, and meanwhile, when the plum blossom-shaped picture is segmented, the iteration number of the CV model is 4 times that of the KLRSF model.
2. The maple leaves with uneven gray scale are segmented, see fig. 3.
The first column is the image to be segmented, the second column is the image obtained by using SR saliency detection, the third column is the result obtained by the present invention, and the fourth column is the result obtained by the CV model. When the maple leaf image is segmented, the iteration number of the CV model is 12 times of that of the KLRSF model. CV models cannot segment models with uneven gray levels and cannot recognize blurred edges.
3. The brain map is shown in fig. 4.
The first column is the image to be segmented, the second column is the image obtained by SR saliency detection, the third column is the result obtained by the present invention, and the fourth column is the result obtained by the CV model. The segmentation effect of the two models is not different in general, and from the details, the islanding phenomenon is more serious when the CV model is segmented than that of the KLRSF model, and it is obvious from the picture that the evolution is stopped because the gray value of the lowest part of the brain in the picture is not greatly different from the gray value of the background and is close to the edge of the picture. The KLRSF model does not completely segment the target object required by the user, and the segmentation effect is only a little better than that of the CV model.
4. The segmentation noise picture is shown in fig. 5.
The first column is the image to be segmented, the second column is the image obtained by using SR saliency detection, the third column is the result obtained by the present invention, and the fourth column is the result obtained by the CV model. Fig. 5 is a picture with gaussian noise, although the noise is high, the influence on the target object is not large, the target object is still clear, the KLRSF model successfully segments the target after 20 iterations, the outer part of the target is not influenced by the noise, the segmented edge is smoother, and the part outside the target object of the CV model is influenced by the noise.
The above disclosure is only a few specific embodiments of the present invention, and those skilled in the art can make various modifications and variations of the present invention without departing from the spirit and scope of the present invention, and it is intended that the present invention also include such modifications and variations as fall within the scope of the appended claims and their equivalents.

Claims (5)

1. An automatic RSF level set image segmentation method based on KL entropy is characterized by comprising the following steps:
acquiring an image to be segmented;
carrying out significance analysis on an image to be segmented, and determining an initial level set of an RSF model;
calculating KL entropy of the image, taking the KL entropy value of the image as a weight coefficient of external energy and internal energy of a segmentation curve, introducing the KL entropy of the image into a level set function of an RSF model, and determining a fitting energy functional of the image to be segmented;
solving an evolution equation according to the fitting energy functional of the image to be segmented, and iteratively solving according to the evolution equation to determine an image segmentation result.
2. The KL entropy based RSF level set image automatic segmentation method according to claim 1, wherein the image to be segmented is subjected to a saliency analysis; the method specifically comprises the following steps: a saliency map is obtained using a residual spectrum method.
3. The KL entropy based RSF level set image automatic segmentation method of claim 1, wherein KL entropy of the image; the method specifically comprises the following steps:
p i to p 0 KL entropy of (a) is expressed as follows:
Figure FDA0002171490820000011
p 0 to p i KL entropy of (a) is expressed as follows:
Figure FDA0002171490820000012
wherein x is a point on the image; p is a radical of formula i And p 0 Probability density distribution functions for the inner region and the outer region, respectively; φ is the level set function of the RSF model.
4. The KL entropy based RSF level set image automatic segmentation method of claim 3, wherein the fitted energy functional of the image to be segmented; the method specifically comprises the following steps:
Figure FDA0002171490820000021
wherein KL (p) i /p 0 ) And KL (p) 0 /p i ) Internal and external energy weight coefficients; f. of 1 (x) And f 2 (x) Respectively, dividing the image gray level mean value inside and outside the curve; h (φ) represents a Heaviside function; i (y) is a given image; k is σ Is a Gaussian function; and v and μ are positive weighting constants.
5. The KL entropy based RSF level set image automatic segmentation method according to claim 4, wherein the evolution equation specifically includes:
Figure FDA0002171490820000022
CN201910764486.9A 2019-08-19 2019-08-19 Automatic RSF level set image segmentation method based on KL entropy Active CN110473224B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910764486.9A CN110473224B (en) 2019-08-19 2019-08-19 Automatic RSF level set image segmentation method based on KL entropy

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910764486.9A CN110473224B (en) 2019-08-19 2019-08-19 Automatic RSF level set image segmentation method based on KL entropy

Publications (2)

Publication Number Publication Date
CN110473224A CN110473224A (en) 2019-11-19
CN110473224B true CN110473224B (en) 2022-09-27

Family

ID=68511077

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910764486.9A Active CN110473224B (en) 2019-08-19 2019-08-19 Automatic RSF level set image segmentation method based on KL entropy

Country Status (1)

Country Link
CN (1) CN110473224B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111693961B (en) * 2020-06-15 2023-05-16 哈尔滨工业大学 CFAR detector based on KL divergence unit screening
CN112508913A (en) * 2020-12-10 2021-03-16 国网江西省电力有限公司电力科学研究院 Cable section edge detection method based on image detection
CN113159969A (en) * 2021-05-17 2021-07-23 广州故新智能科技有限责任公司 Financial long text rechecking system

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109472792A (en) * 2018-10-29 2019-03-15 石家庄学院 In conjunction with the local energy functional of local entropy and the image partition method of non-convex regular terms

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7889941B2 (en) * 2006-11-07 2011-02-15 Siemens Medical Solutions Usa, Inc. Efficient segmentation of piecewise smooth images

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109472792A (en) * 2018-10-29 2019-03-15 石家庄学院 In conjunction with the local energy functional of local entropy and the image partition method of non-convex regular terms

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于局部熵拟合能量与全局信息的活动轮廓模型;王燕等;《计算机工程》;20181225(第12期);全文 *

Also Published As

Publication number Publication date
CN110473224A (en) 2019-11-19

Similar Documents

Publication Publication Date Title
CN109522908B (en) Image significance detection method based on region label fusion
CN110473224B (en) Automatic RSF level set image segmentation method based on KL entropy
CN110837768B (en) Online detection and identification method for rare animal protection
CN109978848B (en) Method for detecting hard exudation in fundus image based on multi-light-source color constancy model
CN111340824B (en) Image feature segmentation method based on data mining
CN112579823B (en) Video abstract generation method and system based on feature fusion and incremental sliding window
CN106157330B (en) Visual tracking method based on target joint appearance model
US11037299B2 (en) Region merging image segmentation algorithm based on boundary extraction
CN114067147B (en) Ship target confirmation method based on local shape matching
Fathi et al. General rotation-invariant local binary patterns operator with application to blood vessel detection in retinal images
Parida et al. Fuzzy clustering based transition region extraction for image segmentation
CN111223063A (en) Finger vein image NLM denoising method based on texture features and binuclear function
CN115690086A (en) Object-based high-resolution remote sensing image change detection method and system
CN107564008A (en) Rapid SAR image segmentation method based on crucial pixel fuzzy clustering
CN112258536B (en) Integrated positioning and segmentation method for calluses and cerebellum earthworm parts
CN110211106B (en) Mean shift SAR image coastline detection method based on segmented Sigmoid bandwidth
CN109241865B (en) Vehicle detection segmentation algorithm under weak contrast traffic scene
CN114596320A (en) Image segmentation method and device based on ALSLCV model
Toure et al. Coastline detection using fusion of over segmentation and distance regularization level set evolution
Bao et al. Solar panel segmentation under low contrast condition
CN110910392B (en) Image segmentation method based on fuzzy clustering of multiple spatial information weighting
Vazquez et al. Image segmentation in the presence of shadows and highlights
Agrawal et al. A Review on Segmentation of Vitiligo image
Li et al. An improved mean shift segmentation method of high-resolution remote sensing image based on LBP and canny features
Wang A new approach for robust segmentation of the noisy or textured images

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