CN106228552A - Gray level image rectangular histogram fast partition method based on mediation K mean cluster - Google Patents
Gray level image rectangular histogram fast partition method based on mediation K mean cluster Download PDFInfo
- Publication number
- CN106228552A CN106228552A CN201610573819.6A CN201610573819A CN106228552A CN 106228552 A CN106228552 A CN 106228552A CN 201610573819 A CN201610573819 A CN 201610573819A CN 106228552 A CN106228552 A CN 106228552A
- Authority
- CN
- China
- Prior art keywords
- image
- cluster
- gray level
- algorithm
- segmentation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of gray level image rectangular histogram fast partition method based on mediation K mean cluster, including input gray level image, be calculated gray level histogram；Determine the cluster segmentation number m of image, mediation K average power exponent p value is set；Initialize subordinated-degree matrix；According to KHM cluster principle, calculate new membership function value matrix U(t+1)；Calculate image weighting function vector W；It is calculated new cluster centre vector Cj；If | | U(t+1)‑U(t)| | < ε or t reaches the greatest iteration value allowed, then algorithm stops iteration；Otherwise t=t+1, proceeds algorithm iteration；The subordinated-degree matrix U obtained according to convergence(t)Image is implemented segmentation.The present invention is insensitive to cluster initial value, and stability is strong, improves the segmentation quality of gray level image, and time efficiency is high, can realize Real-time segmentation task, it is adaptable to the task place that requirement of real-time is high.
Description
Technical field
The present invention relates to the technical field of image segmentation in computer vision, be specifically related to a kind of poly-based on mediation K average
The gray level image rectangular histogram fast partition method of class.
Background technology
At computer vision field, image segmentation is the basis realizing effectively analyzing picture material and understanding, it
All it is widely used at aspects such as industrial practice, modern agricultural production, medical diagnosiss, the surface defects detection of such as workpiece,
Hemocyte segmentation etc. in fruit quality based on image detection, medical microscopic images.Due to the difference of imaging technique, and from
So complexity of boundary's image, to realize the segmentation of accurate image in concrete application scenarios is a job the most difficult, because of
Image partition method has been carried out widely studied for different application demands, Chinese scholars by this, and propose various based on
The effective image dividing method of different background knowledge.
Image Segmentation Technology based on clustering technique is always the study hotspot in image segmentation field.Based on L2-norm
Central cluster technology-K average (K-means, the KM) algorithm of littleization is a famous clustering method for image segmentation.At this
In algorithm, each pixel in image is subordinated to the degree of membership of certain class if it were not for 1, otherwise is just 0, and each pixel is clustered
From the apoplexy due to endogenous wind that certain central pixel point is nearest, therefore KM algorithm is a kind of hard clustering algorithm.Because KM algorithm realizes simple, this algorithm
It is widely used in including that image is segmented in interior various clustering problem.The one of KM algorithm is big, and deficiency is that this algorithm is to cluster centre
Point initial value is very sensitive, and cluster result is unstable, and in cluster process, Chang Yi is absorbed in local extremum solution；It addition, because KM algorithm is
A kind of hard clustering method, therefore in its most disposable clustering problem, certain element i.e. belongs to class A, belongs to again class B or other class simultaneously
Problem.The problem of multiclass may be subordinated to for solving identity element in KM algorithm simultaneously, introduce the mode of dynamic weighting, Dunn
Proposing fuzzy C-mean algorithm (fuzzy C-means, FCM) algorithm, Bezdek is made that improvement to this algorithm thereafter.Propose from FCM
After, this algorithm is also widely used for image segmentation problem.In original FCM algorithm, FCM algorithm is by minimizing in cluster
The Euclidean distance of the heart thus realize cluster process, but this algorithm is the most sensitive to the initial value of cluster centre, and this also causes FCM
Cluster result is unstable.For solution tradition clustering technique based on center to initial value sensitive issue, in recent years, by minimizing all
The mediation distance of value weighting, Zhang et al. proposes mediation K average (K-harmonic means, KHM) clustering algorithm.
Owing to view data in general scale is the biggest, it is the digital picture of M × N for a width size, its pixel count
For M × N, directly image is implemented to split its time efficiency by application KM or FCM algorithm is low-down, and this is to requirement of real-time
The application making these algorithms is very limited by higher industrial practice image processing tasks.Based on this consideration, domestic
Scholar Ye Xiu waits clearly and applies FCM algorithm to achieve the Fast Segmentation to image on the basis of gray level image rectangular histogram, by experiment
Also verifying this fast algorithm and directly applying FCM algorithm is the same in the result that image itself obtains.
On the brace foundation of above technical know-how, in order to improve accuracy and the real-time of image segmentation further,
Image gray levels histogram space application mediation K means clustering algorithm, it is achieved accurate, fly-cutting image.
Summary of the invention
It is an object of the invention to the deficiency overcoming tradition technology based on central cluster in image segmentation, at gray-scale map
As histogram space application mediation K means clustering algorithm realizes carrying out image a kind of new method of Fast Segmentation, with tradition base
Image partition method in central cluster technology is compared, and the present invention can realize the segmentation to image more accurately, and increases result
Stability.
For reaching above-mentioned purpose, the present invention is achieved through the following technical solutions:
Gray level image rectangular histogram fast partition method based on mediation K mean cluster, comprises the steps:
(1) input size is the gray level image of M × N, passes through formula hi=ni/ (M × N) is calculated normalized gradation of image
Level rectangular histogram H={h0,…,hi,…,hL-1, n hereiRepresenting that gray level is the pixel count of i in image to be split, L-1 represents figure
As interior maximum gray scale number, L=256 for 8 bit digital images；
(2) determining the cluster segmentation number m of image, m is cluster centre number here, initializes shape such asC={c 1, c 2, …, c m }
Cluster centre vector, here 0 <c 1<c 2<…<c m <L-1；Mediation K average power exponent p value is set；
(3) make algorithm iteration statistical variable t=0, initialize gray level i be under the jurisdiction of withc j Centered by degree of membership U=[u ij ]L×mSquare
Battle array, in matrix U element meet 0≤u ij ≤ 1 and；
(4) according to KHM cluster principle, according to formulaCalculate new membership function value matrix U(t+1), here | | i-cj| | represent that gray level i is to cluster centrec jEuclidean distance value, p > 0 represent be in harmonious proportion K mean distance power refer to
Number；
(5) according to formulaCalculate the image gray levels i weighting function vector W to each cluster；
(6) according to formula, it is calculated new cluster centre vector Cj；
(7) if | | U(t+1)-U(t)| | < ε or t reaches the greatest iteration value allowed, then algorithm stops iteration, here | |. | | represent
The L of matrix2-norm, ε is a minimum set in advance, and ε value is 1.0e-6 in the present invention；Otherwise t=t+1, forwards to
(4) step proceeds algorithm iteration；
(8) the subordinated-degree matrix U obtained according to convergence(t)Image is implemented segmentation, and exports segmentation result.
Compared with prior art, the present invention applies mediation K means clustering algorithm to carry out in gray level image rectangular histogram quickly
Segmentation, insensitive to cluster initial value, stability is strong, improves the segmentation quality of gray level image, and time efficiency is high, can realize
Real-time segmentation task, it is adaptable to the task place that requirement of real-time is high.
Accompanying drawing explanation
Fig. 1 is the FB(flow block) of the present invention；
Fig. 2 be KM algorithm, FCM algorithm, KHM algorithm based on image gray levels histogram information to bacteria image implement cluster
The result compared that segmentation obtains；Wherein, a is bacteria original image, and b is the knot that KM algorithm carries out that cluster segmentation obtains
Really, c is the result that FCM algorithm carries out that cluster segmentation obtains, and d is that KHM algorithm of the present invention carries out the result that cluster segmentation obtains；
Fig. 3 be KM algorithm, FCM algorithm, KHM algorithm based on image gray levels histogram information to infrared pedestrian's image
Pedestrian image implements the result compared that cluster segmentation obtains；Wherein, a is pedestrian original image, and b is KM
Algorithm carries out the result that cluster segmentation obtains, and c is the result that FCM algorithm carries out that cluster segmentation obtains, and d is KHM algorithm of the present invention
Carry out the result that cluster segmentation obtains；
Fig. 4 be KM algorithm, FCM algorithm, KHM algorithm based on image gray levels histogram information to brain MRI image implement cluster
The result compared that segmentation obtains；Wherein, a is brain MRI original image, and b is the knot that KM algorithm carries out that cluster segmentation obtains
Really, c is the result that FCM algorithm carries out that cluster segmentation obtains, and d is that KHM algorithm of the present invention carries out the result that cluster segmentation obtains.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with instantiation, and with reference to attached
Figure, elaborates to the detailed description of the invention of the present invention, and the present invention is including but not limited to example.
As it is shown in figure 1, be the overall flow figure of the present invention, specifically comprise the following steps that
The first step: input size is the gray level image of M × N, passes through formula hi=ni/ (M × N) is calculated normalized image
Gray level histogram H={h0,…,hi,…,hL-1, n hereiRepresent that gray level is the pixel count of i in image to be split, L-1 table
Maximum gray scale number in diagram picture, L=256 for 8 bit digital images；
Second step, determines the cluster segmentation number m of image, and m is cluster centre number here, initializes shape such asC={c 1, c 2, …,c m Cluster centre vector, here 0 <c 1<c 2<…<c m <L-1, mediation K average power exponent p value is set；
3rd step: the initial value t=0 of Predistribution Algorithm Iterative statistical variable t, initialize gray level i be under the jurisdiction of withc j Centered by be subordinate to
Degree U=[uij]L×mMatrix, in matrix U element meet 0≤u ij ≤ 1 and；
4th step: according to KHM cluster principle, applies formulaIt is calculated new degree of membership letter
Numerical matrix U(t+1), here | | i-cj| | represent that gray level i is to cluster centrec j Euclidean distance value, p > 0 represent be in harmonious proportion K average
Distance power exponent；
5th step: according to formulaCalculate the image gray levels i weighting function vector to each cluster
W；
6th step: according to formulaIt is calculated new cluster centre vector Cj；
7th step: if | | U(t+1)-U(t)| | < ε or t reaches the greatest iteration value allowed, then algorithm stops iteration, here | |. | |
The L of representing matrix2-norm, ε is a minimum set in advance, and ε value is 1.0e-6 in the present invention；Otherwise t=t+1, turns
Algorithm iteration is proceeded to the 4th step；
8th step: obtain subordinated-degree matrix U according to algorithmic statement(t)Image is implemented segmentation and exports segmentation result.
For illustrating effect of the present invention, can be further illustrated by following experiment:
1) experiment condition
Experiment simulation environment is: a CPU is Intel (R) Core (TM) 2 Duo CPU T8100@2.10GHz, operation system
System is Window XP, and programmed environment is the notebook computer of MATLAB R2007b；Experimental image is: a width antibacterial micro-image
Bacteria, an infrared pedestrian image pedestrian, a width is for the MRI image of medical diagnosis；This three width image big
Little by respectively 178 × 178,320 × 240 and 500 × 749；This three width image such as Fig. 2, shown in Fig. 3, Fig. 4, wherein figure (2a) is
Bacteria artwork, figure (3a) is pedestrian artwork, and figure (4a) is MRI image artwork；The tune of the inventive method in experiment
It is set to 3 with K average power exponent p.
2) experiment content
By the present invention and tri-kinds of methods of KM and FCM, three width test image bacteria, pedestrian and MRI image are carried out
Experiment, three kinds of cluster segmentation methods and resultses are such as shown in Fig. 2, Fig. 3 and Fig. 4, and wherein figure (2a) is bacteria original image, figure
(3a) being pedestrian original image, figure (4a) is MRI image original image；Figure (2b), schemes (3b), and figure (4b) is existing KM
Algorithm carries out, to three width tests, the result that cluster segmentation obtains；Figure (2c), schemes (3c), and figure (4c) is that existing FCM algorithm is to three width
Test carries out the result that cluster segmentation obtains；Figure (2d), schemes (3d), and figure (4d) is that three width tests are clustered by the inventive method
The result that segmentation obtains.
3) interpretation
From the segmentation result of Fig. 2, Fig. 3 and Fig. 4 it can be seen that either KM algorithm, or FCM algorithm, at the segmentation knot obtained
All there is place not fully up to expectations more or less in Guo, image object can not preferably be separated from background；As
In the segmentation to bacteria image, KM algorithm and FCM arithmetic result also remain more background noise pixel, especially
Being KM algorithm, in the segmentation to pedestrian and MRI image, the result that FCM algorithm obtains is in cut zone concordance
Poor；The result cut zone inner homogeneous that the inventive method obtains, profile boundary accurate, it is better than the existing cluster compared
The result that partitioning algorithm obtains；It addition, same piece image is being repeated several times in experiment, KM algorithm and FCM algorithm obtain
Convergence result not always consistent, sometimes converge on local extremum, and apply the inventive method to implement to be repeated several times to image
, there is not this phenomenon in experiment, and this also illustrates that the convergence good stability of the present invention is calculated in the traditional KM algorithm compared and FCM
Method.
Table 1 give algorithms of different to bacteria, pedestrian and MRI tri-width test image carry out cluster segmentation time
Calculating time performance compare.
Table 1. image clustering based on gray level histogram information Performance comparision sliced time (unit: second)
As it can be seen from table 1 the calculating time of the inventive method is slightly above FCM algorithm, but much smaller than KM algorithm required time, can
For the image processing tasks that requirement of real-time is higher.
Claims (2)
1. gray level image rectangular histogram fast partition method based on mediation K mean cluster, comprises the steps:
(1) input size is the gray level image of M × N, passes through formula hi=ni/ (M × N) is calculated normalized image gray levels
Rectangular histogram H={h0,…,hi,…,hL-1, n hereiRepresenting that gray level is the pixel count of i in image to be split, L-1 represents image
Interior maximum gray scale number；
(2) determining the cluster segmentation number m of image, m is cluster centre number here, initializes shape such asC={c 1, c 2, …, c m }
Cluster centre vector, here 0 <c 1<c 2<…<c m <L-1；Mediation K average power exponent p value is set；
(3) make algorithm iteration statistical variable t=0, initialize gray level i be under the jurisdiction of withc j Centered by degree of membership U=[u ij ]L×mSquare
Battle array, in matrix U element meet 0≤u ij ≤ 1 and；
(4) according to KHM cluster principle, according to formulaCalculate new membership function value matrix U(t+1), here | | i-cj| | represent that gray level i is to cluster centrec jEuclidean distance value, p > 0 represent be in harmonious proportion K mean distance power refer to
Number；
(5) according to formulaCalculate the image gray levels i weighting function vector W to each cluster；
(6) according to formula, it is calculated new cluster centre vector Cj；
(7) if | | U(t+1)-U(t)| | < ε or t reaches the greatest iteration value allowed, then algorithm stops iteration, here | |. | | represent square
The L of battle array2-norm, ε is a minimum set in advance, and ε value is 1.0e-6 in the present invention；Otherwise t=t+1, forwards (4th) to
Step proceeds algorithm iteration；
(8) the subordinated-degree matrix U obtained according to convergence(t)Image is implemented segmentation, and exports segmentation result.
Gray level image rectangular histogram fast partition method based on mediation K mean cluster the most according to claim 1, its feature
Being, the value of described mediation K mean distance power exponent p is 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610573819.6A CN106228552A (en) | 2016-07-20 | 2016-07-20 | Gray level image rectangular histogram fast partition method based on mediation K mean cluster |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610573819.6A CN106228552A (en) | 2016-07-20 | 2016-07-20 | Gray level image rectangular histogram fast partition method based on mediation K mean cluster |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106228552A true CN106228552A (en) | 2016-12-14 |
Family
ID=57531487
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610573819.6A Pending CN106228552A (en) | 2016-07-20 | 2016-07-20 | Gray level image rectangular histogram fast partition method based on mediation K mean cluster |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106228552A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108919363A (en) * | 2018-08-07 | 2018-11-30 | 哈尔滨工业大学 | One kind is adaptively according to course Aeromagnetic data processing method |
| CN111325658A (en) * | 2020-02-19 | 2020-06-23 | 成都大学 | Color image self-adaptive decolorizing method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030219157A1 (en) * | 2002-05-27 | 2003-11-27 | Tetsushi Koide | Image segmentation method, image segmentation apparatus, image processing method, and image processing apparatus |
| CN101719278A (en) * | 2009-12-21 | 2010-06-02 | 西安电子科技大学 | Automatic tracking method for video microimage cells based on KHM algorithm |
| CN102663681A (en) * | 2012-03-11 | 2012-09-12 | 西安电子科技大学 | Gray scale image segmentation method based on sequencing K-mean algorithm |
| KR20120124094A (en) * | 2011-05-03 | 2012-11-13 | 울산대학교 산학협력단 | Method for clustering image And computer-readable recording medium |
| CN103366367A (en) * | 2013-06-19 | 2013-10-23 | 西安电子科技大学 | Pixel number clustering-based fuzzy C-average value gray level image splitting method |
-
2016
- 2016-07-20 CN CN201610573819.6A patent/CN106228552A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030219157A1 (en) * | 2002-05-27 | 2003-11-27 | Tetsushi Koide | Image segmentation method, image segmentation apparatus, image processing method, and image processing apparatus |
| CN101719278A (en) * | 2009-12-21 | 2010-06-02 | 西安电子科技大学 | Automatic tracking method for video microimage cells based on KHM algorithm |
| KR20120124094A (en) * | 2011-05-03 | 2012-11-13 | 울산대학교 산학협력단 | Method for clustering image And computer-readable recording medium |
| CN102663681A (en) * | 2012-03-11 | 2012-09-12 | 西安电子科技大学 | Gray scale image segmentation method based on sequencing K-mean algorithm |
| CN103366367A (en) * | 2013-06-19 | 2013-10-23 | 西安电子科技大学 | Pixel number clustering-based fuzzy C-average value gray level image splitting method |
Non-Patent Citations (3)
| Title |
|---|
| 叶秀清: "快速模糊分割算法", 《模式识别与人工智能》 * |
| 易生贺: "结合蚁群算法的调和k均值聚类算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 李伟: "改进的基于模糊C－均值聚类的图像分割算法", 《哈尔滨商业大学学报(自然科学版)》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108919363A (en) * | 2018-08-07 | 2018-11-30 | 哈尔滨工业大学 | One kind is adaptively according to course Aeromagnetic data processing method |
| CN108919363B (en) * | 2018-08-07 | 2019-09-13 | 哈尔滨工业大学 | One kind is adaptively according to course Aeromagnetic data processing method |
| CN111325658A (en) * | 2020-02-19 | 2020-06-23 | 成都大学 | Color image self-adaptive decolorizing method |
| CN111325658B (en) * | 2020-02-19 | 2022-08-23 | 成都大学 | Color image self-adaptive decolorizing method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yuan et al. | Factorization-based texture segmentation | |
| Martin et al. | Learning to detect natural image boundaries using local brightness, color, and texture cues | |
| CN107203785A (en) | Multipath Gaussian kernel Fuzzy c-Means Clustering Algorithm | |
| CN105069774B (en) | The Target Segmentation method of optimization is cut based on multi-instance learning and figure | |
| CN103839269A (en) | Image segmentation method based on quaternion and fuzzy C-means clustering | |
| Sonawane et al. | A brief survey on image segmentation methods | |
| CN111950406A (en) | Finger vein identification method, device and storage medium | |
| CN109583493A (en) | A kind of credit card detection and digit recognition method based on deep learning | |
| CN107123130A (en) | Kernel correlation filtering target tracking method based on superpixel and hybrid hash | |
| CN109740631A (en) | Object-based OBIA-SVM-CNN Remote Image Classification | |
| CN108629297A (en) | A kind of remote sensing images cloud detection method of optic based on spatial domain natural scene statistics | |
| CN111815640B (en) | Memristor-based RBF neural network medical image segmentation algorithm | |
| CN105205816A (en) | Method for extracting high-resolution SAR image building zone through multi-feature weighted fusion | |
| CN111539910B (en) | Rust area detection method and terminal equipment | |
| CN107392211B (en) | Salient target detection method based on visual sparse cognition | |
| CN106599927B (en) | The Target cluster dividing method divided based on Fuzzy ART | |
| CN106228552A (en) | Gray level image rectangular histogram fast partition method based on mediation K mean cluster | |
| CN108846845B (en) | SAR image segmentation method based on thumbnail and hierarchical fuzzy clustering | |
| CN108921853B (en) | Image segmentation method based on super-pixel and immune sparse spectral clustering | |
| Huang et al. | Edge detection in UAV remote sensing images using the method integrating Zernike moments with clustering algorithms | |
| CN107423771A (en) | A kind of two phase method for detecting change of remote sensing image | |
| Rahman et al. | Contextual-based top-down saliency feature weighting for target detection | |
| Qian et al. | Weighted superpixel segmentation | |
| CN104123538B (en) | A kind of bad image detecting method of network of view-based access control model bag of words | |
| CN108182684B (en) | Image segmentation method and device based on weighted kernel function fuzzy clustering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161214 |
|
| RJ01 | Rejection of invention patent application after publication |